NEWSMAX

Yeah. So I think for folks who haven't read our website yet, we have this new way of doing data collection, which is building gloves that are mirroring the robot's hand. So instead of needing to deploy thousands and thousands of robots, we just need to make all these gloves and people can wear them and collect data in their own homes. So this gives us really high quality data, but also really high diversity and quantity of data. So I think this year we're going to scale to a few thousands of these people to be clocking data for us every day, and we're going to build a high quality and diverse data set that will be kind of the powering the foundational model that we're going to train. Is there.

Wow. Yeah. To push through that. Yeah. And so, and so at, at this point you're not just dragging that file into a consumer chat bot. At this point you're, you're starting to build custom pipelines. Correct, correct. So I again, I use, I use ChatGPT, I use Gemini, and I use GRO. Yeah. Constantly switching between the two and yeah, built out the pipeline to essentially go through the steps of computational pipeline to get to the mutation. We need to see what's causing the cancer, the root cause. Did you actually use AlphaFold? Is there an open source package that you can download and run? Yeah, we use AlphaFold too. Okay. So from the literature and from also additional LLM sessions, I find out that there's a gene called CKIT that is one of the primary drivers for Rosie's cancer. And what we essentially did was take her healthy DNA. So we sequenced her healthy DNA and we sequenced her cancer DNA, compared them side by side, got like a genetic diff between the two and, and then focused in on the SECA gene, pulled it out, modeled it in AlphaFold. And I used two different techniques to essentially look for drugs to try block the cancer. One was genetic algorithms. So ran genetic algorithms and we actually came up with a unique chemical compound that could block it. But the reason I didn't pursue that is because I actually talked to a chemist about having it made. But the problem with that is you have to go through the steps of first doing it in a test tube, then moving to mouse models and moving on further. It's sort of too complicated. And yeah, the other technique was docking. We docked a whole bunch of these chemical compounds called ligands through the alphafold 3D protein, 3D structure of C kit and mutated CKIT and essentially discovered a drug that was very, very strong at blocking it. But unfortunately the drug is owned by a major US international company. I reached out to them for compassionate use and sure, they politely decline, which is fair enough, but that was kind of the. There's a second part of AlphaFold we used late in the pipeline, but that is kind of the start of the journey. And around this time was about June 2025. And I went through all of that and it really took the wind out of my sails because I tried everything. I tried to see if I could synthesize it, I tried to see if I could get hold of a pre existing chemical. And yeah, one day I was walking Rosie down the street and I realized maybe I'm actually close to making a vaccine myself. And got back on chatgpt and typed away and it said, yeah, you're halfway there. You've already done the DNA sequencing. These are the next steps you need to do. That's amazing. So back to the lab. You did mention we at this point. So I imagine you've looped in friends, colleagues. Who is around you on this project at this point? It's myself and I run a small AI consulting firm here in Sydney. So just I kind of worked it in part time for about two hours every day. Wow. Yeah, it's remarkable. So back to the lab and they wind up finally making the drug. Yeah. So that was a process in of itself went through and did the design of the vaccine construct and pushed. I literally emailed it over to the MRNA Institute at the unsw. It was like half a page of text and the major blocker was actually getting an ethics approval. Because you can't just like go and make MRNA vaccine in your garage. They don't let you do that in Australia. So I've been notified that I had to create an ethics approval and again, I spent, I don't know, that was three months of my life creating that. And it got to a point where we were actually going to have to modify the university's license with the government because the vaccine was going to be administered off site. So the ethics approval would have only been approved in June this year. So. Oh, wow. Through a connection in America, in Seattle, I was connected to Professor Mary May, I don't know how to pronounce his surname. And she is like the preeminent canine cancer person on planet Earth. She connected me to someone in a professor in Queensland, which is a state that's about 1000km, I'm not sure what that is, in miles north of here. I was chatting to her and then I was just saying, like, I'm having trouble with ethics approvals. And she said, oh, actually I have an ethics approval with the government for that specific type of novel immunotherapies and I'm happy to take you under my wing. And I just played it completely cool. I was like, oh, yeah, cool. But I actually like jumping up and down. That's amazing. Yeah. Inside my head. That's remarkable. Oh, that's so cool. So, yeah, so once we got the green light to do that, it all sort of like lined up in parallel. I drove Rose up to Queensland, we did the induction phase of the vaccine and then just sort of waited to see the results. Yeah, Cancer is like a long fight, but it seems like there's at least some really positive signals, something like a 50% decrease in the size of the tumors. Is that roughly correct? How are you measuring progress these days? Okay, there's been a lot of talk about that. So obviously the visual, the best trait is the reduction in the cancer size. We also took blood work, which is going to be published in a paper later this year, and just continuing to visually monitor her tumors, essentially. Yeah, that's great.

Before and you talked to that. You're like this, okay, this is a person. Microsoft Excel 1985 AGI Jose Macedo says ultimate narrative violation from the Dylan Patel Door Cash Pod. Three years later, H1 hundreds are actually trading above launch price in secondary markets. That is negative depreciation. Yeah, that's called appreciation. Appreciation. I appreciate. I just want to go out and say I appreciate age 100 negative depreciation. That's funny. This completely flips a Michael Burry 2 year e waste bear thesis on its head. Yeah, I mean somebody's got to check on. Somebody's got to check on Michael Burry. It's such a different, it's such a different dynamic because I mean like the whole. There was a reasonable underpinning for GPU depreciation which was just look at 20 years of computer equipment history. It's like it all depreciates over like maybe five years, maybe 10 years. Some stuff sticks around, but like they burn out. Yeah, it's just interesting. Jose says Corey probably benefits most from this. They have 250, 50,000 GPUs and a $66 billion backlog. Depending where you think market was pricing, depreciation margins improved by something around 40%, which means 1 billion a year in additional earnings. Who knows where this stuff actually re rates or how sustainable. But great, great time to be a neolab. One of, one of the founding team members at Lambda was posting last week. Basically congratulations to everybody that booked out like GPUs on an annual basis in 2025. You're looking like absolutely brilliant right now. Obviously Sam is starting to look extremely vindicated on all of the deals that he did last year. So Tamaz, quickly let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB don't just build AI.

Admirable. I love it. And I hope that many, many more people hear about this story. I'm sure you've been I'm sure some, like documentary crews and things like that have reached out, but it's really, really special. Well, thank you so much for taking the time. It's great to meet you. Keep, keep us posted. Send us your progress when you when you put out the paper later this year, come back on. And we're sending our prayers to Rosie. Yes. And there we go. Little air horn. Air horn Perot. Well, thank you so much. Incredible stuff to come chat with us. Yeah. Great to meet you, Paul. Have a great rest of your day. Cheers.

Okay. And then. So when do you actually first go to AI tooling? Do you start at a very high level, just sort of asking about dog cancer broadly? Like, at what level did you come into the conversation with AI, just understanding the capabilities? Well, I knew about AlphaFold from the AlphaGo days, so it was the progress, the progression, technology. And I just decided to ChatGPT one day in November 2024, come up with a plan on how we can potentially make a drug to block this cancer. I didn't really know anything about cancer at this stage. I was just going through the process of trying to figure it all out. What happens next? Who do you actually call to? Because at some point it is just text in a box in an app or a website. At what point do you need to go back into the real world to advance the next step? I imagine that ChatGPT at one point tells you, okay, well, we'll need the DNA sequence, and we can't get that just from a text box. So where do you go next? Yes, correct. The first actual piece of data we needed was the DNA sequencing. And, yeah, ChatGPT recommended to reach out to Professor Martin at UNSW. It provided three other people that it was like, it gave all the reason that this is the reason why you should reach out to Professor Martin. That's remarkable. And through a mutual friend here in Sydney, I was connected to Professor Martin and he was very receptive to just taking it on. Extremely receptive, yeah. And so at some point,

It doesn't follow Timothee, Chalamet, et cetera, et cetera. I think what is happening is he came out with like this new, like, it's okay to pursue greatness. Yeah. On the path to greatness. I'm trying to be the goat. I'm trying to, you know, like, coming out with this kind of, like, bravado. Yeah. And if you do that and it's like, me, me, me, me, me, me. I'm trying to be the greatest. Yeah. And then you start just randomly taking shots at another art form where other people are pursuing greatness. Sure. You just invite a lot of criticism because I think, like, everyone's okay, I think, with somebody like, you know, being on their own personal pursuit of greatness. But if you're doing that while trying to tear down other art forms. Yeah. You're just going to invite massive criticism. Yeah. It does feel like he's sort of collapsing, like market cap and Tamil.

You weren't in the states for that project. That's fun. Yeah. Correct. Yeah, that's great. Very cool. And so, yeah, take us through the story of. I actually lost this in the story. Like, when did you. When did you find out that your dog was suffering from cancer? What was the initial process? At what point did you leave the traditional veterinarian system? Sure. So what actually happened? The pre story was Rose had some, like, skin rashes appear on his skin. I took to the vet, and he misdiagnosed it for three times for about 11 months. So over a period of 11 months, took it to the vet, was misdiagnosed, and on the third time, it started bleeding. So I decided to have the tumors removed, and that's when it came back as cancer, unfortunately. Tried really hard to have additional surgery just to remove as much cancer as possible and to, like, essentially try to look at the stem. And because it had been misdiagnosed for so long, one of the tumors that got so large that it wrapped around her leg and we weren't on. There's just not enough skin to close it. Oh. So that's when I kind of realized we needed to do. Try different options. Sure. And. And then tried put it on chemotherapy. But none of the traditional stuff was essentially, like, stopping it. It was continuing to grow. Okay. Okay. And then so.

Marc Andreessen chimes in, I can't load the post right now. We gotta go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32 inch MacBook called the MacBook Pro Ultra Wide, and it looked like this. I bought one and unlocked extreme productivity. And then it wouldn't fit into my backpack, so I had to leave it behind. Oh, no. This is sort of like that other laptop that we saw. They should honestly make this. They should. This should. I mean, walking around looking like maybe you could put skateboard trucks on it. Yeah. That you could use it as transport. Yeah, it's more of like a snowboard build that you like carry over your shoulder like this. Or surfboard. You know, people throw it on the top of your car like that, three fingers. Why you don't put a surfboard on the top of your car? Yeah, I mean, real ones don't. Oh, what do they do? They put it inside the car? Truck batter. Inside truck. Car. Okay. Yeah. I don't pretend in the LA area, you can clock if somebody's actually a surfer or not just by the way they go. Each with their board. No, but I think, yeah. Throwing it under your arm, having some trucks on it, skating, being able to get where you need to go. I like the Ultra Wide. What if they're driving a Huracan Sterrato? Where would you recommend that they put their surfboard then? Jordy Stirato. I can make exceptions. Okay. I like this.

Talk about that transition and how it's evolved. Yeah, yeah. I mean, I think up until 2019, Epic was just media business and that's it. And it would be Google Ads, it'd be YouTube ads, and maybe some brand stuff here and there. And I think in 2019 we did, out of just that pool, a quarter million in revenue. And then that was the year I decided to do product. And so the whole logic being I can't really control any of those three streams of income. Like traffic goes down for one reason or another, all of those go down commensurately. And, and so I thought, okay, well what, what can I sell? And the beauty of having content is that you kind of get like a pre validation engine for, for what you might want to put out there. And so there was this raised bed that I had. It's just like a garden, a metal garden bed that had been sent to me and I was like, this is the thing I get asked the most about. So I'll figure out how to sell it. I didn't even know who gave it to me initially. So I tracked down the manufacturers, Australian company and I just kept emailing them every quarter. I was like, can I sell this? Can I sell this? They said, no, no, no, no, no. They eventually said, yes. I think I had 70 grand in the business, 40 on a shipping container. I knew nothing about E comm. So what I thought I would do is this is the most crazy, stupid E comm logic of all time. But what I thought I would do is bring it into the port of San Diego, which does not take containers. So that, that was already a no go. It goes into the port of Long Beach. I thought I was going to go up and get it. Like me at the port, driving, driving the container down. I have a container here, I'm just picking up. Yeah, just like hauling it down. And then I was looking into Costco Self Storage to like rent that, unload the container and like get like some sort of satellite Internet to print the orders. And I talked to a couple of friends and they were like, yeah, have you heard of a third party logistics company, just ship it there. And you know, just so stupid. But that's how little I knew at the time. And so what happened is made the order, got it on the water, made an Instagram story and said, hey, all these beds you guys keep asking about, they're here Now I have 550 of them. They sold out in two days. Use that cash to buy another container. Sold out that out in two days. So like, by the end of the year, I think we did quarter mil in just that. So the business doubled. And then, of course, setting that up before the global pandemic was insane. So we went from 500 to like 2.8 million to 7.1 million the next year. And then raise, raise a Series A. But, yeah, I mean, immediately I was like, oh, this is obviously the actual revenue driver behind this business, at least. Which I agree, like, a lot of media businesses don't have that easy plugin. Yep, totally. Yeah. What was the team like?

Byrne Hobart, Funny post here. ASML can't figure out how to make money from EUV machines so they sell them tsmc. But TSMC can't figure out how to make money from chips so they sell them to Apple. Apple can't figure out a profitable way to use iPhones so they sell them and there you go the profit. And anyways, Dr. Kareem Carr, is someone saying bear posting. Yes, bear posting that they don't know how to make money from AI directly. This is really, this is such a funny criticism. A funny criticism because if they were, if they, if they actually were, I know exactly where you're going with this. The criticism would be insane. It would be like they created super intelligence and they're keeping it to themselves. Yeah, exactly. The whole point is that every single person on earth, whether you pay for a plan or not. Yeah. Can benefit from today's models. Indeed. Well let's hope.

Volta. Go back to Volta the yeah I mean I remember I was digging into that like chips versus energy. What's the big bottleneck? And I think we're using something like 50% of leading edge capacity like of the. Of the fab of the fabs that can make AI powered GPUs like GPUs that can run transformer based large language models. We're using like 50% of that capacity already. And then some of the leading edge nodes go towards like you know Apple silicon chips that are maybe designed system on a chip, something for a phone and only like 1% of energy right now in America goes towards AI and or less it's like 0.1 or something so you can reallocate and everyone just turn off your air conditioning. One more close the door if the air conditioning lip Bhutan says there no there's no relief as far as I know. No relief until 2028. Somewhat ominous. Keep reading. What how Tomas says what happens when your AI doesn't answer? Everything is in short supply. It's no longer just GPUs it's power, data centers, memory, CPUs. If there's no relief for six more quarters, perhaps it's time to plan for a world where inference isn't freely available on demand. Inference prices which have been static will rise. Subsidies will be harder to justify. Enterprises will need to rationalize workloads deciding which teams receive state of the art models and which don't. Not every CRM update requires a trillion parameter frontier model. Inference rationing normalizes market marketing receives this much sales receives that much. Software engineers probably receive a lot more. Constraint will be the mother of invention. Companies will optimize what they have, adopt open source where they can and likely move to smaller models for many workloads. It's a really cool take. I like this. It's also interesting to me is that not every CRM update requires a trillion parameter frontier model. That that's another bull case for Hopper price stability going forward and less depreciation. Because if you can leave your CRM update workload where you're just going through and spell checking names and cross referencing data sources and pulling from email and dumping some notes summarizing. And you can do that on a GPT4 class model instead of using 5.4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and so that's still economically valuable and so you're able to continue that. Should we go over Ben Thompson's post from this Morning. Yeah, we should now be a good time. Yeah. First, let me tell you about Label Box. RL Environments, Voice Robotics, Evals, and that's where human data. Label Box is the data factory behind the world's leading AI teams. And let me tell you about Vibe Co, where DTC brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audiences and measure sales. Just like on Meta Agents. Published this this morning. To me, the second I saw that I started reading it. It felt like taking a double scoop of C4. Is that a pre workout? Yeah, I know the can. I didn't know it was a. You never dabbled. What was the one that we. I'm more on the gorilla mind one. That's the one that I do. Many people have said you have the mind of a gorilla. Yes, yes, yes. For more plays, more dates, you're a gorilla in sheep's clothing. I think that's literally the pre workout that I have, although I don't use that that often anyway. So you got pumped up. I got pumped up. Ben writes, there's a weird paradox in terms of AI prognostication. Prognostication. Prognostication. That was a good, good effort, Jordy. On one hand, what are. There's just so many words requirements for having a podcast. Like knowing how to say words. No taste. I mean, yeah, ultimately there's a lot of words that you. When you read them. Yeah. You're just like, oh, yeah, you can just do it. And then you try to rip it. On one hand, you don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out to be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I very. I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026. However, on the morning of Nvidia GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gun going. He writes. LLM paradigms over the last couple of weeks, first in the context of Nvidia earnings, and then last week in the context of oracles. I've talked to about 3 LM inflection. I've talked about 3 LM inflection points. Yeah, I'm not going to go through all of these. He goes chat. We've talked about this a few times. LLMs, reasoning models and then agents. And each one of those increases the demand exponentially for computer. So ChatGPT01 and then Opus, as well as Claude Code and Codex. Codex basically getting to the point where tasks are being accomplished over hours and getting to great outcomes. And this is the interesting point, the decreased need for agency. The reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the models so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of capex expenditure not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that computer and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they are going to be used even more than even reasoning models in a chatbot. It's how this third point will be manifested that I think is underappreciated. After all, far more people use chatbots than agents, but I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to use AI. And he goes into a little bit talking about local compute, talking about how Apple's opportunity to run LLMs locally. There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599, I think 499 for education, potentially very disruptive to other laptop makers. You said you still get discounts, Tyler, or does it? I think I'm still stabbed. Oh, yeah, you're still. I'm on leave because you're on leave. That's great. There you go. There are some legendary leave of absences where people have been away for like 10 years and then they go and do so many. See, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before while you're still eligible. That's a good one. Because they're like, oh, well, we got it. I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year, and I think they gave him the honorary degree a couple years later. So, you know, that's the speedrun to beat. But the point about the MacBook Neo is that at 599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with like Asus or another brand? I'm not in the Apple category. Like, it's not an option. Because that store over there, those, those laptops start over 1,000. That's not my budget. So I'm not even going in that store. Well, now you can. And you can spend $600 and get a pretty good computer. And the CFO Nick Wu of ASUS was on their recent ear and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments in the second half of last year. We made some internal prep, but now that it's out, we don't think it's that big of a deal. It has some limitations. Specifically, it only has 8 gigs of RAM, so this is more focused on content consumption. It's not a mainstream notebook for notebook usage, for creation, for working. It's not a work device. It's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, that's what people use these laptops for now. It is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more. And so. And then he goes on to apply that to AI, talking about enterprise and the value of. Companies have a demonstrated willingness to pay for software that makes their employees more productive and AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the pro, not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, basically my interpretation, he's making a case that there are companies that could cut headcount and actually just grow faster. Yeah, if they're implementing AI properly, not just replacing, like the routine workloads. Yeah, so he says. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay, actually, I'm going to start one paragraph. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through huge apparatus filled with humans who accelerate the effort. In some vectors and others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard to manage and motivate human cogs in the organizational machine with agents that not only do what they are told, but do so tirelessly and continuously until the job is done. This only makes the argument that we are not in a bubble much more compelling. Unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously. I'll get around to it when I feel like it. I'll give it a crack. I'll get around. Yeah. This only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of LLMs are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing. Right. You have one. Tyler just, you know, he's going to. Tyler's going to set up to be able to do sign language with his agents to just be. Not even speaking, just sending. Did you actually ever use any of the voice models? Remember Carpathy was talking about that? Yeah. A lot of people do this because you can just like talk much faster, I guess. I haven't done this really. I've used it sometimes use the voice mode, but I don't have to use it in coding agents yet. I was using the ChatGPT voice mode. Like the true back and fourth voice mode. Yeah, like real time voice. Real time voice mode in the car this morning. And they improve that thing dramatically. It's good. Yeah, it's so much better. So first off, it doesn't do that, like that's a great question or anything like that. Or it does that. That whole pause that was in the super bowl ad, like, that just doesn't exist anymore. It just answers and it answers in these like really short, punchy things. I was asking it about, like, how many jobs are actually in America? And it just says like 164 million. And it just like gets me the answer. And I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more detailed back and forth without needing to like dictate a whole prompt and then let pro cook on it for 10 minutes, come back, have it read it. To me, it was like a much better experience. I was very pleasantly surprised by everything, how the back and forth worked. And they also changed it so that you see the floating bubble of like the little animation, but the text populates in real time with the. With your question and then the answer and then your question and the answer. So you can just scroll and read as well. It's very cool. Anyway, third, the last argument that we are not in a bubble. That economic returns from using agents aren't just impactful on the bottom line, that is saving on cost, but the top line as well. In this context, it is any wonder that every single hyperscaler says the demand for compute exceeds supply and that every single hyperscaler is, in the face of stock market skepticism, announcing capex plans that blow away expectations. So I encourage you to go subscribe to Strategy, max out your plan, pay annual. But this was extremely notable. It's such a funny ending where he has this point about like, you only need to be worried about a bubble when you don't need to be worried about a bubble. If everyone's saying a bubble. Because then everyone's like risk off. Because everyone agrees that, oh, we're in a bubble, let's not do bubble behavior. And so capitulation is the sign of a bubble. And he's like, I understand that. And still, this is my take. It's a bold take, but I think it's a good one. Really quickly, let me tell you about the New York Stock Exchange. Want to change the world, raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets. Take us through it, Jordan. He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets. You always want to be, he says. I literally think all the marks are wrong, Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives of the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar, and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary, according to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public. I don't know. Also detailed calling, calling people in private markets arrogant, it's crazy. I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold, bold call by him. But we'll see what evidence he has to back up that extraordinary claim. He blamed the media for creating a frenzy private credit. Obviously we're in the middle of a private credit party. Apparently. If you do credit well, it's honestly, I would say we don't understand private credit well enough to like really put, put everyone up into a frenzy, he says. If you do credit well, it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. Zito talked about the sell off in shares of large software companies, which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity, many with private credit loans, could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good so all is good. Anyone who says that clearly doesn't understand. Most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public yet smaller than those companies and we're trading at a much higher valuation than those companies. And so I am concerned about many of those take privates. Yeah I hear a lot of like Logan Bartlett was doing a ton of analysis in the end of the ZURP era at how high the multiples were in the public markets and that's what was driving the hundred XR transactions. You have to imagine that even, even if we we were like oh yeah that that you know VC backed company was sort of over overhyped at 100xrr. Well that still has a trickle down effect to you know the private equity buyout that's just like yeah. Remember last year when Figma went out. Yeah. And they priced it. Yeah. Very reasonably. Right. They were very intelligent how they priced it. But then obviously there was so much excitement. It's such a great company ran up when it in the first couple days there was. There were some late stage private SaaS companies that I remember were posting like maybe I should go public. Yeah I think it was the Parker rippling was like oh if I can get. That's a crazy multiple. Yeah if I can get you know some insane revenue revenue multiple maybe this appealing. Obviously a lot of those names still could get out this year. Strong companies. They're not not as as eager to get out. Zito pointed To Thoma Bravo's 2021 6.4 billion take private of the software firm Medallia. In particular several lenders to Medallion including Apollo have already written down its debt. He says there will be an issue with respect to that credit which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic small or mid sized Joe Software company. Zito said Joe Software company if he's in the wrong place I think is going to recover somewhere between 20 and 40 cents so 60 to 80% markdown. A lot of the private credit firms have been they'll mark down alone but like mark it down to like 95. Sure you know sure like you know nothing very significant. Zito noted that he expects private credit loans originated in the next 12 to 18 months to be a much better vintage as it relates to quality of company, amount of leverage, documentation and spread. He also weighed in on redemptions and whether private credit managers should enforce limits, typically 5% of a fund's shares each quarter or allow more investors to cash out when they are flooded with requests. It is a topic he and others on Wall street have recently been asked about. As funds take different approaches, you're going to see elevated redemptions for a handful of quarters. I don't know how long it lasts. Making a decision in one quarter may be the right decision for fundraising in the near term, and then a quarter later you'll realize it was a really bad decision. So my overall bias is to stick to the 5% to protect all of my existing investors on vulnerabilities in private equity. Zito sought to shift the focus to private equity, where Apollo has less exposure than most of its peers. He suggested investors voracious demand for buying stakes and existing private equity investments. But wariness of the private debt underpinning those deals doesn't add up since the equity would be junior to the debt if there were major problems with these assets. There's unlimited demand for secondary private equity, but they're worried about private credit, which finances 80% of those portfolios. I can't compute, but I'm the dumb guy. I don't understand. I start saying this and I get these blank stares back at me like, okay, I don't know. He said, I literally think all the marks are wrong. Is that what you're asking me? I think private equity marks are wrong. And again, I read into this. He's talking so candidly, at least in private equity land, he doesn't feel exposed enough to be freaking out. Yeah, a couple more quotes. He says this next cycle is going to be a big moment in time for the private markets because people are way smarter than I think private market participants, particularly people in the wealth channel. Like, I kind of sense an arrogance of the people who grew up in the private markets business. If you don't mark your book, I think you actually lose trust with the clients. We're going to be the market leader in actually marking our book. Let's give it up for being the market leader on the economy and markets. He said, I think it's more likely than not that we go into a recession, a consumer confidence led recession. Most of the companies we lend to are getting a lot of pressure to show clear AI execution. It's forcing people to do stuff before the actual technology works. That's going to be the first step of just a slowdown in the broader economy. Interesting, he said. I literally think Powell, he's so Upset at how it's ending that he's just saying there's inflation every day to piss off the president. Like I literally think that's what's going on. And so hard for me to see inflation. I don't see it anywhere. I see it much more deflationary. I think the technology is attacking every profit pool. What do you say? Asked why a popular high yield high yield corporate bond ETF seen as a benchmark for such debt that is typically under pressure in an economic crunch was relatively flat for the year. Says I don't have any idea. The amount of dispersion going on beneath the surface is kind of crazy. I literally at home, I told my wife last night I feel like the market should be down at least 10% and it's flat or up. Can't make heads or tails of it. On Apollo's credit business, he says on our balance sheet we are 95% investment grade, private and public investment grade. I have a view that bigger companies are going to do better than smaller companies. And so I've tried to position my. The terminology gets me every time. I know because is there like you asked me, I run a private credit fund. We mostly back. We mostly invest in non investment grade opportunities. Yeah, it's like brother, don't you want to be. What were you doing? It's in the name now of course. Very end of this journal is they have a form. We want to hear from you. Are you currently an investor in private credit funds or are you planning to become one? We'd like to hear from you, share your thoughts or experiences in the form below. They're looking for snitches. Yeah. I'm going gig along back to data center land. Amazon. Let me tell everyone about console. Console builds AI agents that automate 70% of it. HR and finance support, giving employees instant resolution for access requests and password resets. And let me also tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. Gotta give another shout out to George Kurtz, who went 1 and 2 again at the Chinese GP over the weekend. Mercedes on an absolute tear. It's the Kurtz effect. I saw some sort of promotional post for a vintage Le mans Racing Series. So 24 hours, but there's some date where like all the cars have to be from early before 1990 or something like that. I don't exactly know how old. I didn't dig into it but it looked very, very cool. Anyway, Cerebrus just landed aws Amazon announced inference chips deal with Cerebras, which is big. They are proving the doubters wrong. Elon is saying that the terrafab project launches in seven days. Ben says what? Very, very fast timeline. Obviously, when people heard about his plans on Doriche, a lot of people kind of questioned it, but Elon's used to being questioned. Yeah. Cerebras is such a cool company. Just the first time. I mean, we've seen it with like the chat Jimmy AI and just going to Codex Desktop, which is of course like a coding harness. But you can just ask it questions and you can Experience Codex. Five point. I think 5.3 is on. Yeah, 5.3. Spark. Spark. Yeah, spark. And it just gives you the answer immediately. And it's actually very, very magical. And I think that's going to be really good for retention. Basically everyone's going to be in the smiling curve. Will smile more as people come back. Math. Zeitlin says data center capacity growth is slowing. He's pulling data here that says newly added US data center capacity slows down considerably in Q4 2025 as market struggles to get keep up with explosive demand. 25 gigawatts of data center capacity added to the funnel in Q4, 50% less in Q3. We'll see if that ramps up again or. That still seems like a lot like the number that I was hearing was for this year. The target for anthropic is like 5 gigawatts, which is like an insane amount of compute. But at the same time, like in the context of 25 gigawatts and one quarter, like, it feels like it. Like there is, like, still significant growth. But of course, you know, risks to all of this. One dozen over on X says they were right to take cigarette ads off tv. I would have smoked a pack a day if I saw this when I was 14. What is this video? Pull it up. Is this a real ad? This cannot be a real ad. I think it's some sort of vibe. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that. Actually, I'm so confused by this ad. I think that's Charlie Sheen. Right? I think. Is that the arc de trio paris? Good music, though. This should be the new launch video. Meta. Oh, it was an international ad. The message there is that they're taking New York to France, but then it was Japanese text on screen. I don't know. It seems like some sort of mashup. I don't know. Let's go over to Tyler Cowen. Yes. With how to lose the AI he wants you to lock in. Oh no, I was going to go to his why you should work much harder right now. Okay, over on Marginal Revolution before we get into the next piece, he says if strong AI will lower the value of your human capital, your current wage is relatively high compared to your future wage. That is an argument for working harder now at least if your current and pending pay can rise with greater effort. Not true for all jobs. If strong AI can at least potentially boost the value of your human capital, you should be investing in learning AI skills right now. No need to fall behind on something so important. You also might have the chance to use that money and buy into the property into the proper capital and land assets. So work harder. He should have put this into a course. I would. I would follow this advice if it cost me $999 in six installments. Yeah, but because he's giving away it for free, it can't. Can't actually be that valuable. Kidding. Of course. From Ricardo in the comments. Suppose you are the best maker of horse carriages in Belgium. Around the time the automobile is invented, you might want to take on as many orders as possible for new carriages because you know future is precarious. Or maybe you get your hands on one of these newfangled automobiles as soon as possible and learn how to fix them. Both options require you to work harder, but these seem to be the two best options available. Paradoxical but true. That's a good take. I like that. A little bit of a white pill. Never, never a bad idea to work harder. Never a bad idea. Should we go through this? Yes, we should. First let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real time experiences and new value with cis. And let me also tell you about cognition. They are the makers of the AI software. Engineer Devon Crush your backlog with your personal AI engineering team. Where do you want to go next? Jordy? How to lose the AI arms race? Let's do it. So investor Leopold Ascherbrenner is now famous for situational awareness. His essay predicted that major AI companies would end up functionally as part of the government led national security project, possibly even nationalized. Along related lines Nick Economist Noah Smith recently asked a critical question. If AI is a weapon, why don't we regulate it like one? We already know. This is Tyler writing in the Free Press the fp.com we already know that the Pentagon has been using Anthropic's Claude to interpret Collected intelligence data and help plan the attacks in Iran. Advanced AI can also be used for cyber attacks, enemy surveillance and identification, followed by missile or drone attacks. Under most extreme scenarios, which may or may not be realistic, advanced AIs might design bioweapons, disable the nuclear weapons of an adversary by disrupting chains of command, or perhaps design and build a scheme to knock missiles out of the sky. Washington, D.C. is starting to ask very basic questions about where we are, what we are doing here. Anthropic and the Department of Defense are at loggerheads over whether Anthropic's AI should be banned from government work. Senator Bernie Sanders recently raised a broader set of concerns, calling for a moratorium on AI data centers with the intention of slowing down progress in artificial intelligence models. But circa 2026, neither nationalization nor an AI slowdown are feasible strategies for the United States. We need to keep our lead both in military and civilian uses of the tech. And that requires a dynamic private sector building our artificial intelligence models. Our federal government, working through the Manhattan Project, developed and built the first atomic bomb. But the strongest AI models are creatures of the private sector, whether we like that fact or not. Even China, which is far more statistically than the United States, has seen its cutting edge models built by companies, not the government. The top AI models are far too complex and require too much high paid talent, including, including international talent, to be done well by governments. Governments sometimes can succeed in building out massive hardware projects, with the space program being another example. They are very. But there are very few cases of government succeeding with advanced software on a large scale. For that, you need private sector dominance. There is no easy way to switch from that mode of organization, which includes salaries of tens of millions of dollars for top researchers, to a more bureaucratic approach. An attempt to do so would destroy or take down those companies, thus thwarting our standing in this new arms race. The general reality is this. We all benefit from living in an advanced civilization rather than eking out subsistence as our ancestors did. But there is part of this bargain we have tended to ignore or take for granted. Now that human beings have developed advanced technologies, we, the freer and better societies must commit to keeping the technological lead. We have not stayed ahead in every area of tech, but we need to be able to protect ourselves and our allies. It's a good thing that America built an atomic bomb before either Hitler or Stalin did. To the extent you believe AI is important, is important for weaponry and national security. That means we need to keep up the pace of progress. You might Find that a slightly. You might find that a slightly unpleasant thought because even under positive visions of an AI future, it will change our world a good deal. Nevertheless, it is a part of technology, of a technological bargain we have been living with for a long time. Arguably since the widespread deployment of firearms or explosives, we seem to have been lulled into a state of stupor by the long standing technological dominance of the United States after World War II. In essence, we have to fight and win yet another arms race. You can't blame AI for that. You can blame AI for real for that reality if you want. But the reemergence of competitive arms races was inevitable with or without AI. You should redirect your ire toward modern history itself. AI may have accelerated the world's new arms race, but there are many other technologies that could play and yet may yet play a comparable role. Space weapons, anyone? How about lasers? Or new types of hypersonic missiles? At least with AI, the US currently holds them the lead. The creativity behind top AI models plays into our national strengths. And he closes by saying, so today we need an odd complex, an odd and complex mix of not entirely consistent ideologies for the current arms race to go well. How about some tech accelerationism mixed with capitalism and then a prudent technocratic approach to military procurement to make sure those advances serve national security ends. On the precautionary side, we need a dash of 1960s and 70s new left and libertarian anti war ideologies skeptical of Uncle Sam himself. We do not want to become the bad guys. Do you think we can pull that off? The new American Challenge is underway. Inspiring. I like this. There was a lot of back and forth around the anthropic Department of War debate. And Door Cash had a great piece on it and lots of people have chimed in now that like the dust has settled a little bit. And I think this is a good sort of nuanced take. It doesn't boil itself down to a tweet just yet, but I think we are getting somewhere with the different trade offs that are at stake. What do you think, Tyler? Yeah, this is good. I mean I think the whole thing that I basically got to when I wrote like the nationalization thing was that like there's just this pretty big scale, right, of like what actually means to nationalize something. There's like the Manhattan Project which is like, okay, this is like full scale, top down, everything is decided by one person. It goes down the pyramid. And then there's like the very kind of distributed like oh, like intel is that Nationalization, I think I broadly agree. Like, I don't think really, I don't think the Manhattan Project is really the best way to do this. Right. Because if you take like, you know, Tatakan's whole thing is like, you know, state capacity libertarianism, like, is the government like fully capable of continuing this AI progress that we have right now? Like, would us stay in the lead if the whole thing is like, you know, set by the government? Yeah, it's unclear. This is sort of what I was going back and forth with Karpon was like, people have framed this as like a battle between Dario Amade and Pete Hegsa. And I feel like we are a democracy and so I would like more authority to be assigned to the individual American voter for a lot of these things. You have that joke about, like, we gotta talk about it, we gotta just talk about this, like, what are we gonna do about AI? It's like, well, like, we can actually vote on it. You can have a plan and then people can vote for it. And there are a bunch of different ways to exercise political will. And it feels like there, there is a trade off, but we got to a good place with the nuclear weapons one. And I do feel like I, as a voter, I have a very small stake, one of 300,000, you know, I guess 300 million, I guess there's like 160 million people that vote in the national election. But part of the national election is, you know, do you trust this particular person to have the nuclear football, to have their finger? They're going to have their finger on the button, like, well, let that sit with you before you cast your ballot. And it will be a continuation of that like this, this is the person that will decide a policy. So vote according to that. Right. And, and I hope that there's more, more of a, of a understanding that the American voter, the American citizen, does have a huge stake in the AI future. And it's not just the, the like, you know, the high flying personalities that give, you know, speeches and podcast appearances. There is a lot more to the American project than that. Well, there is a lot of news around Taiwan and what might happen over there. We found an interesting Kashi market that sort of tracks just general unrest in Taiwan. So the question is, will the United States issue a level four travel advisory for Taiwan? That of course would be a very, very bad news if that did happen. It's sitting at 46% before 2028, January, January 1, 51% before 2029, and 57% above for 2030 and so this is sort of a way to understand geopolitical risk. Obviously, we hope that this calms down and this market goes to zero. Because you sent me that. You sent me that headline about increased activity around Taiwan. Yeah, some of it was part of. Yes, some of it. Yeah. I think the reason that it, that it triggered really calling me out here, sending you fake news like you actually fell for a viral hoax recently. No, I mean, I looked at it and it was factually true. It was just that the activity had dropped enough. The increased activity looked like a really sharp growth, but it was just kind of normalizing. Oh, okay, okay, Interesting. Well, we are certainly hoping for smooth sailing in the Taiwan Strait. Let me tell you about figma. No matter where your idea starts, figma make clog code codex or sketch. The Figma canvas is where ideas connect and products take shape. Build in the right direction with figma. Asml. Bern Hobart, Funny post here. ASML can't figure out how to make money from EUV machines, so they sell them to tsmc. But TSMC can't figure out how to make money from chips, so they sell them to Apple. Apple can't figure out a profitable way to use iPhones, so they sell them and there you go, the profit. And anyways, Dr. Kareem Carr, is someone saying Bear posting. Yes, Bear posting that they don't know how to make money from AI directly. This is really. This is such a funny criticism. It's such a funny criticism because if they actually were doing with this, the criticism would be insane. It would be like they created super intelligence and they're keeping it to themselves. Yeah, exactly. The whole point is that every single person on Earth, whether you pay for a plan or not, can benefit from today's models. Indeed. Well, let's head over to Meta. But first let me tell you about 11 labs. Build intelligent real time conversational agents. Reimagine human technology interaction with 11 labs. So Nebias and Meta have agreed to a $27 billion AI infrastructure pact deal. The talks are advanced to Pact stage. Five year deal, 27 billion to supply AI infrastructure capacity to Meta. Nebias has really been an unturned fascinating company. Formerly part of Yandex, spun out independent, now publicly traded and just one of the Neo clouds that's figured out that Microsoft deal and now seems to be doing good work with Meta. So Navia said it will provide $12 billion of dedicated capacity across multiple locations. Meta will also purchase up to 15 billion in additional capacity over the five year. Over the five year period. These deals are sort of squishy, but it doesn't matter because the people who actually need to know can underwrite them accordingly. Nebius added that it will use large scale deployments of Nvidia's next generation Vera Rubin AI infrastructure, which Jensen is surely talking about at GTC right now. That's expected to be available in the second half of the year, and Nebias will begin delivery of that capacity beginning early next year, which feels like a decade in AI timelines. Why do you have the paper in front of your face? The team earlier said I look like a third base coach, so I'm covering up. Yeah, because you don't want to let everyone know what play you're calling. There you go. Exactly. There was news Friday, late a rumor or some reporting from Reuters. Metta is planning sweeping layoffs that could affect 20% or more of the company, three sources familiar with the matter told Reuters. As Meta seeks to offset AI infrastructure bets and prepare for greater efficiency brought by AI assisted workers. How many employees does Meta have? I think it's like 60,000, something like that. Let's figure 75. 75,000. June 30, 2025. That's the same number. Yeah. 78,000 as of December 31st. Somewhere in the same range as Salesforce. And again, not super surprising. Stock's up around 2% today. I would expect this to pop even harder once these layoffs are actually announced. Yeah, I mean, the advice is become aligned with the AI effort at Meta. If these layoffs happen, they're clearly cutting part of the workforce. But then they're also acquiring and hiring all over the place, just more around AI. I mean, we saw that today with the Manus announcement. They're taking new naming Meta just call your product a computer. Manus Computer, we got. No, no, no. It's called My Computer Manus. My Computer by Manus. My Computer by Manus. It works on mobile, works on your computer. Manis desktop. But wait again, this was My computer is the core feature of the new Manus desktop app. It's your AI API. Okay, so it's still called Manis. Direct competitor to Codex Claude Code Code Cowork and Microsoft Cowork. At this point, everyone's doing Cowork, so maybe you just rip that. Yeah. So the reason I thought the Manus acquisition was interesting at the time is people were positioning it as more of a talent acquisition. Like, these are great product builders that figured out how to grow products super quickly. I think at the time they sold, they were somewhere in the range of 100 to 200 million of run rate. I was interested in it specifically because it seemed like Zuck was trying to take what they had built and actually just scale it, not just roll them into working on ads or whatever other products. Tyler, please download my computer by Manus and play around with it and come back with a review. So the top recommended action that they showcase here is organizing thousands of unsorted photos. I'm not super into organization for the sake of organization, but that does seem pretty useful. I was taking photos on an actual camera this weekend and had to transfer them from the camera to an iPad, then scroll through them, favorite them, then share them over airdrop. And there is a cool agentic workflow which is basically actually download the raws. Some of them were a little bit overexposed, some of them need a little bit of color grading. And if I could have a workflow where Manus or some desktop agent opens every photo individually in Photoshop and tweaks it and does it intelligently and crops it ever so slightly and is thoughtful about it, that would definitely speed up my life. Dodd says would trust openclaw more than Manus after the meta acquisition with private data. Yeah, well, the Manus branding, the meta branding on this is so limited. I would be surprised if people sort of, you know, if this. If this goes broad, people wouldn't necessarily know that much. I wonder if they'll do the Oculus thing and you'll have to, like, log in with Facebook at some point. You can log into Facebook. You can already, but you can also log in with normal, like email. Yeah. I mean, Manas, before was. Wasn't it a Chinese company? It was based in Singapore, but it was, you know, like, rumored to be aligned with China. And so, I mean, not rumored. They. They were building it in China. Okay. To Singapore, because the optics were not good. So, you know, as far as private data security goes, I think this is an upgrade. Right? It certainly feels like it. Anyway, let me tell you about Phantom cash. Fund your wallet without exchanges or middlemen and spend with the Phantom card. So the Oscars happened last night. Jordi, just to get you up to speed, the Oscars are an award show that are put on by the Motion Picture and Academy of Arts and Sciences. Yeah, I saw someone on the ramp cap table got an award. Yeah, yeah, yeah. Michael B. Jordan won best Actor and he won Best Investor for that. Best Investor. Yeah. They should have a category for that, but Timothee Chalamet is getting taken to task in the Financial Times. Over his views on opera and ballet, of all things. The Financial Times writes, it's quite sweet, really. So desperate are some people to get their knickers in a twist on the Internet that in the face of a lull in the culture wars, we have real wars now. The only thing they have found to get outraged about recently relates to a man saying, nobody cares about ballet and opera anymore. Anymore. The man I refer to as Timothee Chalamet, a talented young actor who stars in the multi Oscar nominated Marty supreme, which had a very unfortunate showing at the Oscars. I think they were nominated for nine awards and they didn't win anything. And so upset is your belief that it had to do with his comments disrespecting? No. Or it was just now the people, the, the critics actually just said, hey, like, you know, yeah, I think in every category he was. Marty supreme was up against like a Goliath. Like it was a. Every fight was sort of a David and Goliath. And there were just no upsets because he was going up against sinners and one battle after another, which were heavy favorites, I think, from the very beginning, before these comments were made. So Timothee Chalamet was talking with a fellow actor, Matthew McConaughey, at a town hall event organized by CNN and Variety in February. But the comments actually just got clipped and went viral recently. It was two week delay. The slicers over there got to step it up. He said, I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive. Even though, like, no one cares about this anymore. All respect to the ballet and opera people out there. And then he said distinctly, disrespectfully, I just lost 14 cents in viewership. Damn. I just took shots for no reason. There is evidence of Chalamet showing having made similar comments before, such as on the Graham Norton show in 2019 when he called opera a quote, outdated art form. And at an event the same year where he was where he was worried that cinema would become like opera or ballet or something, kind of a dying art form or something. He also, as many people, as many of those who claim to feel so offended have pointed out, has close family connections to the world of classical dance. His mother, grandmother and sister all danced with the New York City Ballet. Wow. And he has spoken out about growing up dreaming big backstage at the Koch Theater in New York, where the ballet performs. As someone who tried to pursue a career in pop music, while my older sister. This is the writer in the Financial Times, My older sister pursued one in classical piano. I would wager that he has been honing this particular attack, or perhaps defense line since adolescence. So his apparent instant regret, his slip felt. Felt a bit disingenuous. Are you a. Are you an opera fan? Ballet fan? I like the opera. Me too. Although I actually haven't not been to ballet opera yet, so it's hard for me to. And I just think, like, there's a world where, you know, the film and movie industry, like, does become like opera and ballet, but that's still like a beautiful thing with an amazing. We've seen this in la, where I think a lot of movies are now releasing only at these, like, kind of fancy theaters. Yeah. Tarantino has these kind of things where it's like much more like kind of upstage a real event that you go to. Yeah. And of course it is. It is like, you know, just technological disruption with social media and there's a lot of other, like, gyrations in the transition there. But I'll tell you why I think this whole kerfuffle's happened. Kerfuffle happened. And as someone who doesn't really follow Hollywood, doesn't follow film, doesn't follow Timothee, Chalamet, et cetera, et cetera. I think what is happening is he came out with this new it's okay to pursue greatness on the path to greatness. Sure, sure, sure. I'm trying to be the goat. I'm trying to, you know, like, with this kind of like bravado. Yeah. And if you do that and it's like, me, me, me, me, me, me. Sure. I'm trying to be the greatest. Yeah. And then you start just randomly taking shots at another art form where other people are pursuing greatness. Sure. You just invite a lot of criticism because I think, like, everyone's okay, I think with somebody like, you know, being on their own personal pursuit of greatness. But if you're doing that while trying to tear down other art forms. Yeah. You're just gonna invite massive criticism. Yeah. It does feel like he's sort of. He's sort of collapsing like market cap and like tam of like, yes. The opera tam and the ballet tam is smaller than film, but it would be odd. Play the actual sound. People that are here that are younger than me were people desire are desiring things that are more patient and that pull you in. I just saw another article that says Gen Z is a bigger movie going audience than millennial audience. You know, I feel like a fucking grandpa saying that no, but point being, I think even, like, Frankenstein, which was like, a hugely popular movie this year, I didn't think that pacing was extraordinarily fast or anything, but it pulled people in, you know? But it does take you having to wave a flag of, hey, this is a serious movie or something, and some people want to be entertaining quickly. I'm really right in the middle, Matthew, because I admire people and I've done it myself. To go on a talk show, hey, we got to keep movie theaters alive. We got to keep this genre alive. And another part of me feels like if people want to see it, like Barbie, like Oppenheimer, they're going to go see it and go out of their way to be loud and proud about it. And I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive, even though no one cares about this anymore. All respect to the ballet and opera people out there. I just lost 14 cents in viewership. But crazy shots. That's not a shot. I hear what you're saying. Yeah, yeah, Yeah, yeah. I don't know, it's interesting. I was thinking about, like, if, like, the creator of GTA 5, like, stood on stage and was just like, we are 10 times the size of the baseball. Baseball. But also, like, the movie industry, like the gate. The video gaming industry has been basically 10 times the size of the. Of the movie industry for. You mean the movie theater business? No, like. Like Hollywood, like, gross production. Yeah, totally. I'm almost positive. Not. Not 10 times the size of your streaming platform. Yeah, maybe streaming. That includes TV shows. And then do you include mobile games or not? That's a big question. But the video game industry is definitely bigger. Raghav in the Twitch chat From deep Nvidia CEO just said he sees 1 trillion in revenue through 2020. Bring down the gong. Bring down the male. Congratulations. That's amazing. Thank you. And we have our next guest in the Restream waiting room. First, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And we are joined by Kevin from Epic Gardening in the Restream waiting room. Let's bring in the CB Pro jump. Kevin, how are you doing? What's going on? What's up, brothers? How you doing? Good to see you, brother. Thanks so much for taking the time to join the show. First up, we gotta talk about that tank. Yeah, what's in the tank? What's in the tank we've been talking about? I'm breeding rare Costa Rican tree frogs in this tank. No way. Yeah, they're endangered, okay. Oh, they're endangered. Okay. What else is. What else is special about. And you're planning to release them in all 50 states? States. Once you have enough. This is the goal. This is the goal we're always scaling over here. Yeah, of course. I love it. I love it. Is it challenging? Like, how much of your time is devoted to that particular tank? Almost none. Almost none. I just need to make sure that they're fed. Yeah, that's cool. What do they eat? Yeah, they eat crickets, which I'm breeding in that little tank right over there. You can see that. Yeah, we're breeding the different trophic levels over here, for sure. Okay, and then. And then. Do these frogs have use in your garden? Is it purely just for fun? No, I'm just branching out to flora or to fauna now, I guess. Yeah. Here at Epic, you know. Okay, well, first time in the show, so I want to kick off with your backstory. I want to know about the decision to start making content. I feel like that's always an interesting origin story. Like, when did you think, okay, I need to make content, dude. I mean, I'm an Internet OG, so I was on GeoCities. I was on Angelfire back in the day making, like, anime tutorials, you know, So I don't know what it is. I think genetically I'm designed to make content, but for Epic, it was really a calling card for. Remember when you used to design WordPress websites back in the day? Like, when people actually paid for that service? I used the blog as, like, a calling card or a digital business card for, like, designing websites for local businesses, and then just kind of kept plugging along with it and adding different platforms. And here we are today. Yeah. What about the first YouTube video? Like, what was the backstory behind choosing to go to YouTube? Choosing to go to video. It's a big lift for people if they're on substack or they're a writer and they don't know how they're going to do in front of camera. First YouTube video was 2013. So it was a long time ago. And ironically, back then, I mean, SEO and blogs were kind of the thing. Yeah. And so for me, the first YouTube video. Maybe it's the second YouTube video. You can see me using a screen recording app, reading a blog article. Just literally reading the article. Yeah. And with the hopes that people would watch that video and click the blog link, and I would make money off of the advertising on the blog. So it's a completely backwards logic to today, of course. Yeah, yeah, yeah. But then obviously discover YouTube is a far better platform, especially these days. So what was the. What was the flow of traffic over time? Were you able to reroute blog viewers to YouTube or did the algorithm eventually kick in because you're pre algo feed, right? Yeah, yeah, I think so. Right. I mean, I think it was back then, if you subscribe to a channel on YouTube, that subscription would just show up, which was a beautiful time. But no, I think every platform, as you expand every platform, you think, like, okay, well, I can get someone from this one to that one. It tends not to work. You tend to have to just play each platform for what it is. And so, like, YouTube became its own thing insta. All the other social media platforms have become their own thing. Now. How do you think about, like, serializing content, creating through lines, like the initial formats? Like, what was the actual development of the playbook that you ran on YouTube? On YouTube, I think in the early days, because remember, like, I'm 13 years old as a YouTuber, which is like two YouTuber lifespans, I think last about six years or so. And so back in the early days, it was just pure SEO, especially for a gardening channel. It's like, hey, how do you. How do I grow basil? How do I grow tomatoes? How do I grow.

Is that not every CRM update requires a trillion parameter frontier model. That's another bull case for hopper price stability going forward and less depreciation. Because if you can leave your CRM update workload where you're just going through and spell checking names and cross referencing data sources and pulling from email and dumping some notes, summarizing, and you can do that on a GPT4 class model instead of using five. Four, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and so that's still economically valuable and so you're able to continue that. Should we go over Ben Thompson's post from this morning? Yeah, we should. Now be a good time. First, let me tell you about Label Box, RL environments, Voice robotics evals and expert human data. Label Box is the data factory behind the world's leading AI teams. And let me tell you about Vibe Co, where DTC brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audiences and measure sales. Just like on Meta Agents published this this morning. To me, the second I saw that I started reading it, it felt like taking a double scoop of C4. Is that a pre workout? Yeah. You never. I know the can. I didn't know it was a. You never dabbled. What was the one that was we. I'm more of the gorilla mind one. That's the one that I. Many people have said you have the mind of a gorilla. Yes, yes, yes. For more plates, more dates, you're a gorilla in sheep's clothing. I think that's literally the pre workout that I have, although I don't use it that often anyway. So you got pumped up. I got pumped up. Ben writes, there's a weird paradox in terms of AI prognosticization. Prognostication. Prognostication. That was a good, good effort, Jordan. There's just so many words. What are the requirements for having a podcast? Like knowing how to say words. No taste. I mean, yeah, ultimately there's a lot of words that you, when you read them. Yeah. You're just like, oh yeah, you can just do it. And then you try to rip it on one hand. You don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out to be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former. I've very. I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026, however, on the morning of Nvidia's GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which, paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gun going, he writes. LLM paradigms over the last couple of weeks, first in the context of Nvidia earnings, and then last week in the context of Oracles. I've talked to you about 3lLM inflection. I've talked about 3lM inflection points. I'm not going to go through all of these. He goes chad, we've talked about this a few times. LLMs, reasoning models, and then agents, and each one of those increases the demand exponentially for compute. So LM, ChatGPT01 and then Opus, as well as Claude Code and Codex, basically getting the point where tasks are being accomplished over hours. Yep. And getting to great outcomes. And this is the interesting point. Okay, the Decreased need for agency the reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the model so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of capex expenditure, not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that compute and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they are going to be used even more than even reasoning models in a chatbot. It's how this third point will be manifested that I think is underappreciated. After all, Far more people use chatbots than agents. But I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to use AI. And he goes into a little bit talking about local, local compute, talking about how Apple's opportunity to run LLMs locally. There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599, I think 499 for education, potentially very disruptive to other laptop makers. You still get discounts, Tyler, or does it? I think I'm still scammed. Oh, yeah, you're still scammed because you're on leave. That's great. There you go. There are some legendary leave of absences where people have been away for like 10 years and then they go and do so many. See, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before, while you're still eligible. That's a good one because they're like, oh, well, we got it. I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year, and I think they gave him the honorary degree a couple years later. So, you know, that's the speedrun to beat. But the point about the MacBook Neo is that at 599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with, like, Asus or another brand? I'm not in the Apple category. Like, it's not an option. Because, yeah, that, that store over there, those, those laptops start over 1,000. That's not my budget. So I'm not even going in that store. Well, now you can. And you can spend $600 and get a pretty good computer. And the CFO Nick Wu of Asus was on their recent earnings call and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments in the second half of last year. We made some internal prep, but now that it's out, like, we don't think it's that big of a deal. Like, it has some limitations. Specifically, it only has eight gigs of ram. So, like, you know, this is more focused on content consumption. It's not a Mainstream notebook for notebook usage, for creation, for working. It's not a work device, it's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, that's what people use these laptops for now. It is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more. And then he goes on to apply that to to AI. Talking about enterprise and the value of Companies have a demonstrated willingness to pay for software that makes their employees more productive. And AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the pro, not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, basically my interpretation, he's making the case that there are companies that could cut headcount and actually just grow faster. Yeah, if they're implementing AI properly, not just replacing like the routine workloads. Yep. So he says. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay, actually, I'm going to start one paragraph. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through huge apparatus filled with humans who accelerate the effort in some vectors and retard it in others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard to manage and motivate human cogs in the organizational machine. With agents that not only do what they are told, but do so tirelessly and continuously until the job is done. This only makes the argument that we are not in a bubble much more compelling. First, unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously. I'll. I'll get around to it. When I feel like it, I'll give it a crack. I'll get around. Yeah. This only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of LLMs are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing. Right. You have one. Tyler just, you know, he's going to. Tyler's going to set up to be able to do sign language with his agents to just, just be. Not even speaking. Just, just. Do you actually ever use any of the voice models? Remember Carpath? He was talking about that. Yeah. A lot of people do this because you can just like talk much faster, I guess. I haven't done this really. I've used it sometimes use the voice mode, but I don't actually use it in like coding agents yet. I was using the ChatGPT voice mode. Like the true, like back and forth voice mode. Yeah, like real time voice. Real time voice mode in the car this morning and they improved that thing dramatically. It's good. Yeah, it's so much better. So first off, it, it doesn't do that, like that's a great question or anything like that. Or that whole pause that was in the super bowl ad. Like that just doesn't exist anymore. It just answers and it answers in these like really short, punchy things. I was asking it about like, how many jobs are actually in America and it just says like 164 million. And it just gets me the answer. And I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more jobs detailed back and forth without needing to dictate a whole prompt and then let Pro cook on it for 10 minutes, come back, have it read it. To me, it was a much better experience. I was very pleasantly surprised by how the back and forth worked. And they also changed it so that you see the floating bubble of the little animation, but the text populates in real time with your question and then the answer and then your question and the answer. So you can just scroll and read as well. It was very cool. Anyway, third, the last argument, that we are not in a bubble. That economic returns from using agents aren't just impactful on the bottom line, that is saving on costs, but the top line as well. Let's go in this context, it is any wonder that every single hyperscaler says the demand for compute exceeds supply and that every single hyperscaler is in the face of stock market skepticism, announcing capex plans that blow away expectations. So I encourage you to go subscribe to STRATECHERI max out your plan, pay annual. But this was extremely notable. It's such a funny ending where he, he, he has this point about like, you only need to be worried about a bubble when like, you don't need to be worried about a bubble if everyone's saying a bubble because then, then everyone's like risk off because everyone agrees that we're, oh, we're in a bubble. Let's not do bubble behavior. And so capitulation is, is the sign of a bubble. And he's like, I understand that. And still, this is my take. It's a bold take, but I think it's a good one. Really quickly, let me tell you about the New York Stock Exchange. Want to change the world, Raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets. Take us through it, Jordan. He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets. You always want to be sound, he says. I literally think all the marks are wrong. Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives at the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients. Late last month, Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary. According to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public. I don't know. Also detailed calling, you know, people in private markets arrogant is crazy. I feel like, I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold, bold call by him. But we'll see what he, what evidence he has to back up that extraordina claim. He blamed the media for creating a frenzy around private Credit obviously we're in the middle of a private credit party apparently if you do credit well it's honestly I would say we don't understand private credit well enough to like really put, put everyone up into a frenzy. He says if you do credit well, it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. Zito talked about the sell off in shares of large software companies which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity, many with private credit loans, could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good so all is good. Anyone who says that clearly doesn't understand most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public yet smaller than those companies. And we're trading at a much higher valuation than those companies. And so I am concerned about many of those take privates. Yeah, I remember a lot of like Logan Bartlett was doing a ton of analysis in the end of the ZIRP era at how high the multiples were in the public markets and that's what was driving the hundred x ARR transactions. And you have to imagine that even, even if we were like oh yeah, that VC backed company was sort of overhyped at 100x ARR. Well that still has a trickle down effect to the private equity buyout. That's just like yeah, remember last year when Figma went out and they priced it very reasonably. Right. They were very intelligent how they priced it. But then obviously there was so much excitement because it's such a great company. It ran up in the first couple days. There was, there were some late stage private SaaS companies that I remember were, were posting like maybe I should go public. Yeah, I think it was the, the Parker rippling was like oh if I can get. That's a crazy multiple. Yeah, if I can get you know, some insane revenue revenue multiple maybe this appealing. Obviously a lot of those names still could get out this year. Strong companies. They're not, not as, as eager to get out. Zito pointed To Thoma Bravo's 2021 6.4 billion take private of the software firm Medallia in particular several Lenders to Medallion, including Apollo, have already written down its debt. He says there will be an issue with respect to that credit, which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic small or mid sized Joe Software Co. Zito said, Joe Software Co. If he's in the wrong place, I think he's going to recover somewhere between 20 and 50, 40 cents. So 60 to 80% markdown. A lot of the private credit firms have been they'll mark down alone but like mark it down to like 95. Sure you know, sure you know nothing very significant. Zito noted that he expects private credit loans originated in the next 12 to 18 months to be a much better vintage as it relates to quality of company, amount of leverage, documentation and spread. He also weighed in on redemptions and whether private credit managers should enforce limits, typically 5% of a fund's shares each quarter or allow more investors to cash out when they are flooded with requests. It is a topic he and others on Wall street have recently been asked about. As funds take different approaches, you're going to see elevated redemptions for a handful of quarters. I don't know how long it lasts. Making a decision in one quarter may be the right like decision for fundraising in the near term and then a quarter later you'll realize it was a really bad decision. So my overall bias is to stick to the 5% to protect all of my existing investors on vulnerabilities in private equity. Zito sought to shift the focus to private equity where Apollo has less exposure than most of its peers. He suggested investors voracious demand for buying stakes in existing private equity investments. But wariness of the private debt underpinning those deals doesn't add up since the equity would be junior to the debt if there were major problems with these assets. There's unlimited demand for secondary private equity, but they're worried about private credit, which finances 80% of those portfolios. I can't compute, but I'm the dumb guy. I don't understand. I start saying this and I get these blank stares back at me like, okay, I don't know. He said, I literally think all the marks are wrong. Is that what you're asking me? I think private equity marks are wrong. And again, I read into this. He's talking so candidly, at least in private equity land, he doesn't feel exposed enough to be freaking out. A couple more quotes. He says this next cycle is going to be a big moment in time for the private Markets because people are way smarter than I think private market participants, particularly people in the wealth channel. Like, I kind of sense an arrogance of the people who grew up in the private markets business. If you don't mark your book, I think you actually lose trust with the clients. We're going to be the market leader and actually marking our book. Let's give it up for being the market leader on the economy and markets. He said, I think it's more likely than not that we go into a recession, a consumer confidence led recession. Most of the companies we lend to are getting a lot of pressure to show clear AI execution. It's forcing people to do stuff before the actual technology works. That's going to be the first step of just a slowdown in the broader economy. Interesting. He said. I literally think Powell, he's so upset at how it's ending that he's just saying there's inflation every day to piss off the president. Like I literally think that's what's going on. It's so hard for me to see inflation. I don't see it anywhere. I see it much more deflationary. I think the technology is attacking every profit pool. What do you say? Asked why a popular high yield, high yield corporate bond ETF seen as a benchmark for such debt that is typically under pressure in an economic crunch, was relatively flat for the year, says I don't have any idea. The amount of dispersion going on beneath the surface is kind of crazy. I literally at home, I told my wife last night I feel like the market should be down at least 10% and it's flat or up. Can't make heads or tails of it. On Apollo's credit business, he says on our balance sheet we are 95% investment grade, private and public investment grade. I have a view that bigger companies are going to do better than smaller companies. And so I've tried to position my. The terminology gets me every time. Yeah, I know because is there, like you ask me, I run a private credit fund. We mostly back. We mostly invest in non investment grade opportunities. Yeah, it's like, brother, don't you want to be invested? What were you doing? It's in the name now of course at the end of this journal piece they have a form. We want to hear from you. Are you currently an investor in private credit funds or are you planning to become one? We'd like to hear from you, share your thoughts or experiences in the form below. They're looking for snitches. Yeah, I'm going gig along back to Data Center Land. Amazon. Let me tell everyone about console. Console builds AI agents that automate 70% of it. HR and finance support, giving employees instant resolution for access requests and password resets. And let me also tell you about Crowdstrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. Got to give another shout out to George Kurtz, who went 1 and 2 again at the Chinese GP over the weekend. Mercedes on an absolute tear. It's the Kurtz effect. I saw a. I saw some sort of promotional post for a vintage Le Mans racing series. So 24 hours, but there's some date where, like all the cars have to be from early. Before 1990 or something like that. I don't exactly know how old. I didn't dig into it, but it looked very, very cool. Anyway, Cerebrus just landed aws. Amazon announced Inference chips deal with Cerebras, which is big. They are proving the doubters wrong. Elon is saying that the terrafab project launches in seven days. Beff Jesus says what? Very, very fast timeline. Obviously, when people heard about his plans on Dwark Keshe, a lot of people kind of questioned it, but Elon's used to being questioned. Yeah, Cerebras is such a cool company. Just the first time. I mean, we've seen it with the chat. Jimmy, AI and you know, just going to Codex Desktop, which is of course like a coding harness, but you can just ask it questions and you can experience codecs. Five point. I think 5.3 is on. Yeah, 5.3. Spark. Spark. Yeah, spark. And it just gives you the answer immediately. And it's actually very, very magical. And I think that's going to be really good for retention. Basically everyone's going to be in the smiling curve. Will smile more as people come back. And Matthew Zeitlin says data center capacity growth is slowing. He's pulling data here that says newly added US data center capacity slows down considerably in Q4 2025 as market struggles to keep up with explosive demand. 25 gigawatts of data center capacity added to the funnel in Q4, 50% less in Q3. We'll see if that ramps up again or. That still seems like a lot like the number that I was hearing was for this year. The target for anthropic is like 5 gigawatts, which is like an insane amount of compute. But at the same time, like in the context of 25 gigawatts and one quarter, like, it feels like, like the. There Is like still significant growth. But of course, you know, risks to all of this. One dozen over on X says they were right to take cigarette ads off tv. I would have smoked a pack a day if I saw this when I was 14. What is this video? Pull it up. Is this real ad? This cannot be a real ad. I think it's some sort of vibe. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that actually, I'm so confused by this ad. I think that's Charlie Sheen. Right? Is that the arc de trio paris? Good music, though. This should be the new launch video. Meta. Oh, it was an international ad. The message there is that they're taking New York to France. But then it was Japanese text on screen. I don't know. It seems like some sort of. Some sort of mashup. I don't know. Let's go over to Tyler Cowan. Yes. With how to Lose the AI he wants you to lock in. Oh, no, I was going to go to his. Why you should work much harder now. Okay. Over on Marginal Revolution. Before we get into the next piece, he says if strong AI will lower the value of your human capital, your current wage is relatively high compared to your future wage. That is an argument for working harder now, at least if your current and pending pay can rise with greater effort. Not true for all jobs. If strong AI can at least potentially boost the value of your human capital, you should be investing in learning AI skills right now. No need to fall behind on something so important. You also might have the chance to use that money and buy into the proper. Into the proper capital and land assets. So work harder. He should have put this into a course. I would follow this advice if it cost me $999 in six and six installments. But because he's given away it for free, it can't actually be that valuable. Kidding, of course. From Ricardo in the comments. Suppose you're the best maker of horse carriages in Belgium. Around the time the automobile is invented. You might want to take on as many orders as possible for new carriages because you know your future is precarious. Or maybe you get your hands on one of these newfangled automobiles as soon as possible and learn how to fix them. Both options require you to work harder, but these seem to be the two best options available. Paradoxical but true. That's a good take. I like that. Little bit of a white pill. Never, never a bad idea to work harder. Never a bad idea. Should we go through this? Yes, we should first let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real time experiences and new value with Cisco. And let me also tell you about cognition. They are the makers of the AI software. Engineer Devin Crush your backlog with your personal AI engineering team. Where do you want to go next? Jordy how to lose the AI Arms race let's do it. So investor Leopold Aschenbrenner is now famous for situational awareness. His essay predicted that major AI companies would end up functionally as part of the government led national security project, possibly even nationalized. Along related lines, economist Noah Smith recently asked a critical question. If AI is a weapon, why don't we regulate it like one? We already know this is Tyler writing in the Free Press the fp.com we already know that the Pentagon has been using Anthropic's CLAUDE to interpret collected intelligence data and help plan the attacks in Iran. Advanced AI can also be used for cyber attacks, enemy surveillance and identification, followed by missile or drone attacks. Under most extreme scenarios, which may or may not be realistic, advanced AIs might design bioweapons, disable the nuclear weapons of an adversary by disrupting chains of code command, or perhaps design and build a scheme to knock missiles out of the sky. Washington, D.C. is starting to ask very basic questions about where we are, what we are doing here. Anthropic and the Department of Defense are at loggerheads over whether Anthropic's AI should be banned from government work. Senator Bernie Sanders recently raised a broader set of concerns, calling for a moratorium on AI data centers with the intention of slowing down progress in artificial intelligence models. But circa 2026, neither nationalization nor an AI slowdown are feasible strategies for the United States. We need to keep our lead both in military and civilian uses of the tech, and that requires a dynamic private sector building our artificial intelligence models. Our federal government, working through the Manhattan Project, developed and built the first atomic bomb. But the strongest AI models are creatures of the private sector, whether we like that fact or not. Even China, which is far more statistical than the United States, has seen its cutting edge models built by companies, not the government. The top AI models are far too complex and require too much high paid talent, including international talent, to be done well by governments. Governments sometimes can succeed in building out massive hardware projects, with the space program being another example. But there are very few cases of government succeeding with advanced software on a large scale. For that you need private sector dominance. There is no easy way to switch from that mode of organization, which Includes salaries of tens of millions of dollars for top researchers to a more bureaucratic approach. An attempt to do so would destroy or take down those companies, thus thwarting our standing in this new arms race. The general reality is this. We all benefit from living in an advanced civilization rather than eking out subsistence as our ancestors did. But there is part of this bargain we have tended to ignore or take for granted. Now that human beings have developed advanced technologies, the freer and better societies must commit to keeping the technological lead. We have not stayed ahead in every area of tech, but we need to be able to protect ourselves and our allies. It's a good thing that America built an atomic bomb before either Hitler or Stalin did. To the extent you believe AI is important, is important for weaponry and national security. That means we need to keep up the pace of progress. You might find that a slightly. You might find that a slightly unpleasant thought. Because even under positive visions of an AI future, it will change our world a good deal. Nevertheless, it is a part of technology, of a technological bargain we have been living with for a long time. Arguably since the widespread deployment of firearms or explosives. We seem to have been lulled into a state of stupor by the long standing technological dominance of the United States after World War II. In essence, we have to fight and win yet another arms race. You can't blame AI for that. You can blame AI for that reality if you want, but the reemergence of competitive arms races was inevitable with or without AI. You should redirect your ire toward modern history itself. AI may have accelerated the world's new arms race, but there are many other technologies that could play and yet may yet play a comparable role. Space weapons, anyone? How about lasers or new types of hypersonic missiles? At least with AI, the US currently holds the lead. The creativity behind top AI models plays into our national strengths. And he closes by saying so. Today we need an odd complex, an odd and complex mix of not entirely consistent ideologies for the current arms race to go well. How about some tech accelerationism mixed with capitalism and then a prudent technocratic approach to military procurement to make sure those advances serve national security ends. On the precautionary side, we need a dash of 1960s and 70s new left and libertarian anti war ideologies skeptical of Uncle Sam himself. We do not want to become the bad guys. Do you think we can pull that off? The new American challenge is underway. Inspiring. I like this. There was a lot of back and forth around the anthropic department of war Debate and Door Cash had a great piece on it and lots of people have chimed in now that like, the dust has settled a little bit. And I think this is a, is a good sort of nuanced take. It doesn't, it doesn't boil itself down to a tweet just yet, but I think we are getting somewhere with the different trade offs that are at stake. What do you think, Tyler? Yeah, this is good. I mean, I think the whole thing that I basically got to when I wrote like the nationalization thing was that like, there's just this. There's pretty big scale, right, of like, what actually means to nationalize something. There's like the Manhattan Project, which is like, okay, this is like full scale, top down. Everything is decided by, by one person. It goes down the pyramid and then there's like the very, you know, kind of distributed like. Oh, like intel. Is that nationalization? Yeah, I think, I broadly agree. Like, I don't think really, I don't think the Manhattan Project is really the best way to do this. Right. Because if you take like, you know, total council thing is like, you know, state capacity libertarianism, like, is the government fully capable of continuing this AI progress that we have right now? Would us stay in the lead if the whole thing is set by the government? It's unclear. This is sort of what I was going back and forth with Karpon was people have framed this as a battle between Dario Amadei and Pete Hegseth. And I feel like we are a democracy. And so, like, I would like more, more authority to be assigned to the individual American voter for a lot of these things. You know, you have that joke about, like, we got to talk about it. We got to just talk about this. Like, what are we going to do about AI? It's like, well, like, we can, we can actually vote on it. Like, you can, you can have a plan and then people can vote for it. And there are a bunch of different ways to exercise political will. And it feels like there, there is a trade off, but we got to a good place with the nuclear weapons one. And I do feel like I, as a voter, I have a very small stake, one of 300,000. You know, I guess I have 300 million. I guess there's like 160 million people that vote in the national election. But.

Models and which don't. Not every CRM update requires a trillion parameter frontier model. Inference rationing normalizes market marketing receives this much sales receives that much software engineers probably receive a lot more constraint will be the mother of invention. Companies will optimize what they have, adopt open source where they can and likely move to smaller models for many workloads. I like this. It's also interesting to me is that that not every CRM update requires a trillion parameter frontier model. That that's another bull case for Hopper price stability going forward and less depreciation. Because if you can leave your CRM update workload where you're just going through and spell checking names and cross referencing data sources and pulling from email and dumping some notes, summarizing and you can do that on a GPT4 class model instead of using 5.4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and so that's still economically valuable and so you're able to continue that. Should we go over Ben Thompson's post from this morning? Yeah, we should. Now be a good time. Yeah. First, let me tell you about Labelbox, RL Environments, Voice Robotics evals and expert human data. Label Box is the data factory behind the world's leading AI teams. And let me tell you about Vibe Co, where D2C brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audiences and measure sales. Just like on Meta Agents over Bubbles Agents published this this to me. The second I saw that I started reading it. It felt like taking a double scoop of C4. Is that a pre workout? Yeah. You never. I know the can. I didn't know it was a. You never dabbled. What was the one that we. I'm more of the gorilla mind one. That's the one that I do. Many people have said you have the mind of a gorilla. Yes, yes, yes. From more plays, more dates. You're a gorilla in sheep's clothing. I think that's literally the pre workout that I have. Although I don't use that often anyway. So you got pumped up. I got pumped up. Ben writes, there's a weird paradox in terms of AI prognostication. Prognostication. Prognostication. That was a good, good effort, Jordy. On one hand there's some of these. There's just so many words. What are the requirements for having a podcast? Like knowing how to say words? No taste. I mean, yeah, ultimately there's a lot of words that you, when you read them? Yeah. You're just like, oh yeah, you can just do it. And then you try to rip it on one hand. You don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out to be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I have very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026, however, on the morning of Nvidia's GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gun going. He writes. LLM paradigms over the last couple of weeks, first in the context of Nvidia earnings, and then last week in the context of Oracles. I've talked to you about 3 lm inflection. I've talked about 3 lm inflection points. I'm not going to go through all of these. He goes, chad, we've talked about this a few times. LLMs, reasoning models and then agents. And each one of those increases the demand exponentially for compute. So LM chatgpt01 and then opus as well as Claude Code and Codex. Codex, yeah, basically getting the point where tasks are being accomplished over hours and getting to great outcomes. And this is the interesting point, okay, the decreased need for agency. The reason Ben has been writing about these three inflection points over the last couple weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens, because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the models so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of capex expenditure not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply first, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that computer and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they are going to be used even more than even reasoning models in a chatbot. It's how this third point will be manifested that I think is underappreciated. After all, far more people use chatbots than agents. But I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to use AI. And he goes into a little bit talking about local compute, talking about how Apple's opportunity to run LLMs locally. There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599, I think 499 for education, potentially very disruptive to other laptop makers. You said that you still get discounts, Tyler. Or does. I think I'm still scanned. Oh, yeah, you're still standing. I'm on leave because you're on leave. That's great. There you go. There are some legendary leave of absences where people have been away for like 10 years and then they go and do so many. See, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before, while you're still eligible. That's a good one. Because they're like, oh, well, we got it. I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year, and I think they gave him the honorary degree a couple years later. So, you know, that's the. That's the speedrun to beat. But the point about the MacBook Neo is that at 599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with like Asus or another brand? I'm not in the Apple category. Like, it's not an option because that store over there, those laptops start over 1,000. That's not my budget. So I'm not even going in that store. Well, now you can. And you can spend $600 and get a pretty good computer. And the CFO Nick Wu of Asus was on their recent earning, and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments in the second half of last year. We made some internal prep, but now that it's out, we don't think it's that big of a deal. It has some limitations. Specifically, it only has 8 gigs of RAM. So this is more focused on content consumption. It's not a mainstream notebook for notebook usage, for creation, for working. It's not a work device. It's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, like, that's what people use these laptops for now. It is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more. And so. And then he goes on to apply that to AI, Talking about enterprise and the value of Companies have a demonstrated willingness to pay for software that makes their employees more productive. And AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, basically my interpretation, he's making the case that there are companies that could cut headcount and actually just grow faster. Yeah, if they're implementing AI properly, not just replacing, like the routine workloads. Yep, so he says. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay, actually, I'm going to start one paragraph. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through a huge apparatus filled with humans who accelerate the effort in some vectors and retard it in others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard to manage and motivate human cogs in the organizational machine with agents that not only do what they are told, but do so tirelessly and continuously until the Job is done. This only makes the argument that we are not in a bubble much more compelling. First, unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously. I'll. I'll get around to it. When I feel like it, I'll give it a crack. I'll get around. Yeah. So this only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of LLMs are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing. Right. You have one. Tyler just, you know, he's going to. Tyler's going to set up to be able to do sign language with his agents to just. Just be. Not even speaking. Just. Just send. Did you actually ever use any of the voice models? Remember Carpathy was talking about that? Yeah, A lot of people do this because you can just like talk much faster, I guess. I haven't done this really. I've used it sometimes use the voice mode, but I don't have to use it in like coding agents yet. I was using the ChatGPT voice mode. Like the true, like, back and fourth voice mode. Yeah, like real time voice. Real time voice mode in the car this morning. And they improved that thing dramatically. It's good. Yeah, it's so much better. So first off, it doesn't do that, like that's a great question or anything like that, or that whole pause that was in the super bowl ad. Like, that just doesn't exist anymore. It just answers and it answers in these, like, really short, punchy things. I was asking it about, like, how many jobs are actually in America? And it just says like 164 million. And it just like gets me the answer and I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more detailed back and forth without needing to like, dictate a whole prompt and then let pro cook on it for 10 minutes, come back, have it, read it. To me, it was like a much better experience. I was very pleasantly surprised by how the back and forth worked. And they also changed it so that you see the floating bubble of like, the little animation, but the text populates in real time with the. With your question and then the answer and then your question and the answer. So you can just scroll and read as well. It's very cool. Anyway, third, the last argument that we are not in a bubble. The economic returns from using agents aren't just impactful on the bottom line, that is saving on cost, but the top line as well. In this context, it is any wonder that every single hyperscaler says the demand for compute exceeds supply and that every single hyperscaler is in the face of stock market skepticism, announcing CapEx plans that blow away expectations. So I encourage you to go subscribe to strategy, max out your plan, pay annual. But this was extremely notable. It's such a funny ending where he has this point about like, you only need to be worried about a bubble when you don't need to be worried about a bubble. If everyone's saying a bubble because then everyone's like risk off because everyone agrees that oh, we're in a bubble, let's not do bubble behavior. And so capitulation is the sign of a bubble. And he's like, I understand that. And still this is my take. It's a bold, it's a bold take, but I think it's a good one. Really quickly, let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets. Take us through it, Jordan. He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets. You always want to be, he says. I literally think all the marks are wrong, Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives of the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar, and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary, according to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public I don't know. Also detailed calling, you know, people in private markets arrogant. It's crazy. I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold, bold call by him, but we'll see what he, what evidence he has to back up that extraordinary claim. He blamed the media for creating a frenzy around PR private credit. Obviously we're in the middle of a private credit party. Apparently if you do credit well, it's honestly, I would say we don't understand private credit well enough to like really put everyone up into a frenzy. He says if you do credit well, it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. Zito talked about the sell off in shares of large software companies which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity, many with private credit loans, could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good, so all is good. Anyone who says that clearly doesn't understand. Most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public yet smaller than those companies and we're trading at a much higher valuation than those companies. And so I am concerned about many of those take privates. Yeah. I remember a lot of like Logan Bartlett was doing a ton of analysis in the end of the ZURP era at how high the multiples were in the public markets and that's what was driving the 100XR transactions. And you have to imagine that even, even if we, we were like oh yeah, that, that you know, VC backed company was sort of over, overhyped at 100x RR. Well that still has a trickle down effect to you know, the private equity buyout. That's just like. Yeah. Remember last year when Figma went out? Yeah. And they priced it. Yeah. Very reasonably. Right. They were very intelligent how they priced it. But then obviously there was so much excitement. It's such a great company ran up when it, in the first couple days there was, there was some late stage private SaaS companies that I remember were posting like API should go public. Yeah. I think it was the Parker rippling was like oh if I can get. That's a crazy multiple. Yeah, if I can get, you know some insane revenue revenue multiple maybe this appealing. Obviously a lot of those names still could get out this year. Strong companies. They're not not as as eager to get out. Zito pointed To Thoma Bravo's 2021 6.4 billion take private of the software firm Adalia. In particular, several lenders to Medallia, including Apollo, have already written down its debt. He says there will be an issue with respect to that credit which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic smaller mid sized Joe Software company. Zito said Joe Software company if he's in the wrong place, I think is going to recover somewhere between 20 and 40 cents. So 60 to 80% markdown. A lot of the private credit firms have been they'll mark down alone but like mark it down to like 95. Sure, you know, you know nothing very significant. Zito noted that he expects private credit loans originated in the next 12 to 18 months to be a much better vintage as it relates to quality of company, amount of leverage, documentation and spread. He also weighed in on redemptions and whether private credit managers should enforce limits, typically 5% of a fund's shares each quarter or allow more investors to cash out when they are flooded with requests. It is a topic he and others on Wall street have recently been asked about. As funds take different approaches. You're going to see elevated redemptions for a handful of quarters. I don't know how long it lasts. Making a decision in one quarter may be the right decision for fundraising in the near term and then a quarter later you'll realize it was really bad decision. So my overall bias is to stick to the 5% to protect all of my existing investors on vulnerabilities in private equity. Zito sought to shift the focus to private equity where Apollo has less exposure than most of its peers. He suggested investors voracious demand for buying stakes in existing private equity investments. But wariness of the private debt underpinning those deals doesn't add up since the equity would be junior to the debt if there were major problems with these assets. There's unlimited demand for secondary private equity, but they're worried about private credit which finances 80% of those portfolios. I can't compute but I'm the dumb guy. I don't understand. I start saying this and I get these blank stares back at me like okay, I Don't know. He said, I literally think all the marks are wrong. Is that what you're asking me? I think private equity marks are wrong. And again, I read into this. He's talking so candidly, at least in private equity land, he doesn't feel exposed enough to be freaking out. Yeah, a couple more quotes. He says this next cycle is going to be a big moment in time for the private markets because people are way smarter than I think private market participants, particularly people in the wealth channel. Like, I kind of sense an arrogance of the people who grew up in the private markets business. If you don't mark your book, I think you actually lose trust with the clients. We're going to be the market leader in actually marking our book. Let's give it up for being the market leader on the economy and markets. He said, I think it's more likely than not that we go into a recession, a consumer confidence led recession. Most of the companies we lend to are getting a lot of pressure to show clear AI execution. It's forcing people to do stuff before the actual technology works. That's going to be the first step of just a slowdown in the broader economy. Interesting. He said. I literally think Powell, he's so upset at how it's ending that he's just saying there's inflation every day to piss off the president. Like I literally think that's what's going on. And so hard for me to see inflation. I don't see it anywhere. I see it much more deflationary. I think the technology is attacking every profit pool. What do you say? Asked why a popular high yield corporate bond ETF seen as a benchmark for such debt that is typically under pressure in an economic crunch, was relatively flat for the year. Says I don't have any idea. The amount of dispersion going on beneath the surface is kind of crazy. I literally at home, I told my wife last night I feel like the market should be down at least 10% and it's flat or up. Can't make heads or tails of it. On Apollo's credit business, he says on our balance sheet we are 95% investment grade, private and public investment grade. I have a view that bigger companies are going to do better than smaller companies. And so I've tried to position myself. The terminology gets me every time. I know because is there, like you asked me, I run a private credit fund. We mostly back. We mostly invest in non investment grade opportunities. Yeah, it's like, brother, don't you want to be. What were you doing? It's in the name now, of course. Very end of this journal is they have a form. We want to hear from you. Are you currently an investor in private credit funds or are you planning to become one? We'd like to hear from you, share your thoughts or experiences in the form below. They're looking for snitches. Yeah, I'm going gig along. Back to data center land. Amazon. Let me tell everyone about console. Console builds AI agents that automate 70% of it. HR and finance support, giving employees instant resolution for access requests and password resets. And let me also tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. Got to give another shout out to George Kurtz, who went 1 and 2 again at the Chinese GP over the weekend. Mercedes on an absolute tear. It's the Kurtz effect. I saw a. I saw some sort of promotional post for a vintage Le mans racing series, so 24 hours, but there's some date where, like all the cars have to be from early, before 1990 or something like that. I don't. I don't exactly know how old. I didn't dig into it, but it looked very, very cool. Anyway, Cerebrus just landed. Aws. Amazon announced inference chips deal with Cerebras, which is big. They are proving the doubters wrong. Elon is saying that the Terrafab project launches in seven days. Beff Jesus says what? Very, very fast timeline. Obviously, when people heard about his plans on Dwark Cash, a lot of people kind of questioned it, but Elon's used to being questioned. Yeah. Cerebras is such a cool company. Like just the first time. I mean, we've seen it with like the chat, Jimmy, AI and, you know, just going to Codex Desktop, which is of course like a coding harness, but you can just ask it questions and you can Experience Codex. Five point. I think 5.3 is on. Yeah, 5.3. Spark. Spark. Yeah, spark. And it just gives you the answer immediately. And it's actually very, very magical. And I think that's going to be really good for retention. Basically, everyone's going to be in the smiling curve. Will smile more as people come back. Math. Zeitlin says data center capacity growth is slowing. He's pulling data here that says newly added US data center capacity slows down considerably in Q4 20, 25 as market struggles to keep up with explosive demand. 25 gigawatts of data center capacity added to the funnel in Q4. 50% less in Q3. We'll see if that ramps up again. Or. That still seems like a lot. Like the number that I was hearing was for this year. The target for anthropic is like 5 gigawatts, which is like an insane amount of compute. But at the same time, like, in the context of 25 gigawatts in one quarter, like, it feels like, like there is, like still significant growth. But of course, you know, risks to all of this. One dozen over on X says they were right to take cigarette ads off tv. I would have smoked a pack a day if I saw this when I was 14. What is this video? Pull it up. Is this real ad? This cannot be a real ad. I think it's some sort of vibe. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that actually, I'm so confused by this ad. I think that's Charlie Sheen. Right? Is that the arc de triomphe, paris? Good music, though. This should be the new launch video. Meta. Oh, it was an international ad. The message there is that they're taking New York to France. But then it was Japanese text on screen. I don't know, it seems like some sort of. Some sort of mashup. I don't know. Let's go over to Tyler Cowen. Yes. How to lose the AI he wants you to lock in. Oh, no. I was going to go to his. Why you should work much harder now. Okay. Over on Marginal Revolution. Before we get into the next piece, he says if strong AI will lower the value of your human capital, your current wage is relatively high compared to your future wage. That is an argument for working harder now. At least if your current and pending pay can rise with greater effort. Not true for all jobs. If strong AI can at least potentially boost the value of your human capital, you should be investing in learning AI skills right now. No need to fall behind on something so important. You also might have the chance to use that money and buy into the property, into the proper capital and land assets. So work harder. He should have put this into a course. I would. I would follow this advice if it cost me $999 in six installments. Yeah, but because he's giving away it for free, it can't. Can't actually be that valuable. Kidding. Of course. From Ricardo in the comments. Suppose you are the best maker of horse carriages in Belgium. Around the time the automobile is invented, you might want to take on as many orders as possible for new carriages because, you know, future is precarious. Or maybe you get your hands on one of these newfangled automobiles as soon as possible and learn how to fix them. Both options require you to work harder, but these seem to be the two best options available. Paradoxical but true. That's a good take. I like that. A little bit of a white pill. Never, never a bad idea to work harder. Never a bad idea. Should we go through this? Yes, we should. First let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real time experiences and new value with cis. And let me also tell you about cognition. They are the makers of the AI software Engineer Devon Crush your backlog with your personal AI engineering team. Where do you want to go next? Jordy how to lose the AI arms race? Let's do it. So investor Leopold Ascherbrenner is now famous for situational awareness. His essay predicted that major AI companies would end up functionally as part of the government led national security project, possibly even nationalized. Along related lines, Nicolas economist Noah Smith recently asked a critical question. If AI is a weapon, why don't we regulate it like one? We already know this is Tyler writing in the Free Press the fp.com we already know that the Pentagon has been using Anthropic's Claude to interpret collected intelligence data and help plan the attacks in Iran. Advanced AI can also be used for cyber attacks, enemy surveillance and identification, followed by missile or drone attacks. Under most extreme scenarios, which may or may not be realistic, advanced AIs might design bioweapons, disable the nuclear weapons of an adversary by disrupting chains of command, or perhaps design and build a scheme to knock missiles out of the sky. Washington, D.C. is starting to ask very basic questions about where we are, what we are doing here. Anthropic and the Department of Defense are at loggerheads over whether Anthropic's AI should be banned from government work. Senator Bernie Sanders recently raised a broader set of concerns, calling for a moratorium on AI data centers with the intention of slowing down progress in artificial intelligence models. But circa 2026, neither nationalization nor an AI slowdown are feasible strategies for the United States. We need to keep our lead both in military and civilian uses of the tech, and that requires a dynamic private sector building our artificial intelligence models. Our federal government, working through the Manhattan Project, developed and built the first atomic bomb. But the strongest AI models are creatures of the private sector, whether we like that fact or not. Even China, which is far more statistical than the United States, has seen its cutting edge models built by companies not the government. The top AI models are far too complex and require too much high paid talent, including, including international talent, to be done well by governments. Governments sometimes can succeed in building out massive hardware projects, with the space program being another example. They are very. But there are very few cases of government succeeding with advanced software on a large scale. For that you need private sector dominance. There is no easy way to switch from that mode of organization, which includes salaries of tens of millions of dollars for top researchers, to a more bureaucratic approach. An attempt to do so would destroy or take down those companies, thus thwarting our standing in this new arms race. The general reality is this. We all benefit from living in an advanced civilization rather than eking out subsistence as our ancestors did. But there is part of this bargain we have tended to ignore or take for granted. Now that human beings have developed advanced technologies, the freer and better societies must commit to keeping the technological lead. We have not stayed ahead in every area of tech, but we need to be able to protect ourselves and our allies. It's a good thing that America built an atomic bomb before either Hitler or Stalin did. To the extent you believe AI is important, is important for weaponry and national security. That means we need to keep up the pace of progress. You might find that a slightly. You might find that a slightly unpleasant thought, because even under positive visions of an AI future, it will change our world a good deal. Nevertheless, it is a part of technology, of a technological bargain we have been living with for a long time. Arguably since the widespread deployment of firearms or explosives, we seem to have been lulled into a state of stupor by the long standing technological dominance of the United States after World War II. In essence, we have to fight and win yet another arms race. You can't blame AI for that. You can blame AI for real for that reality if you want. But the reemergence of competitive arms races was inevitable with or without AI. You should redirect your ire toward modern history itself. AI might may have accelerated the world's new arms race, but there are many other technologies that could play and yet may yet play a comparable role. Space weapons, anyone? How about lasers? Are new types of hypersonic missiles? At least with AI, the US currently holds the lead. The creativity behind top AI models plays into our national strengths. And he closes by saying so. Today we need an odd complex, an odd and complex mix of not entirely consistent ideologies for the current arms race to go well. How about some tech accelerationism mixed with capitalism and then a prudent technocratic approach to military procurement to make sure those advances serve national security ends. On the precautionary side, we need a dash of 1960s and 70s new left and libertarian anti war ideologies skeptical of Uncle Sam himself. We do not want to become the bad guys. Do you think we can pull that off? The new American challenge is underway. Inspiring. I like this. There was a lot of back and forth around the anthropic Department of War debate and Door Cash had a great piece on it and lots of people have chimed in now that like the dust has settled a little bit. And I think this is a good sort of nuanced take. It doesn't boil itself down to a tweet just yet, but I think we are getting somewhere with the different trade offs that are at stake. What do you think, Tyler? Yeah, this is good. I mean, I think the whole thing that I basically got to when I wrote like the nationalization thing was that like, there's just this. There's pretty big scale, right, of like, what actually means to nationalize something. There's like the Manhattan Project, which is like, okay, this is like full scale, top down. Everything is decided by, by one person. It goes down the pyramid. And then there's like the very, you know, kind of distributed, like, oh, like intel. Is that nationalization? Yeah, I think, I broadly agree. Like, I don't think really, I don't think the Manhattan Project is really the best way to do this right. Because if you take like, you know, total council thing is like, you know, state capacity libertarianism, like, is the government fully capable of continuing this AI progress that we have right now? Would us stay in the lead if the whole thing is set by the government? It's unclear. This is sort of what I was going back and forth with Karp on was people have framed this as a battle between Dario Amadei and Pete Hegseth. And I feel like we are a democracy and so, like, I would like more, more authority to be assigned to the individual American voter for a lot of these things. You know, you have that joke about, like, we got to talk about it, we got to just talk about this. Like, what are we going to do about AI? It's like, well, like, we can, we can actually vote on it. Like, you can, you can have a plan and then people can vote for it. And there are a bunch of different ways to exercise political will. And it feels like there, there is a trade off, but we got to a good place with the nuclear weapons one. And I do feel like I, as a voter I have a very small stake, one of 300,000. You know, I guess 300 million, I guess there's like 160 million people that vote in the national election. But part of the national election is, you know, do you trust this particular person to have the nuclear football, to have their finger? They're going to have their finger on the button like, well, let that sit with you before you cast your ballot. And it will be a continuation of that. Like this, this is the person that will decide AI policy. So vote according to that. Right. And I hope that there's more, more of a, of a understanding that the American voter, the American citizen does have a huge stake in the AI future. And it's not just the, the like, you know, the high flying personalities that give, you know, speeches and podcast appearances. There is a lot more to the American project than that. Well, there is a lot of news around Taiwan and what might happen over there. We found an interesting Kalxi market that sort of tracks just general unrest in Taiwan. So the question is, will the United States issue a level four travel advisory for Taiwan? That of course would be a very, very bad news if that did happen. It's sitting at 46% before 2028, January, January 1, 51% before 2029, and 57% above for 2030. And so this is sort of a way to understand geopolitical risk. Obviously, we hope this calms down and this market goes to zero. Because you sent me that, you sent me that headline about increased activity around Taiwan. Yeah, some of it was part of it. Yeah, some of it. I think the reason that it, that it triggered really calling me out here, sending you fake news. You're like, you actually fell for a viral hoax recently. No, I mean, I looked at it and it was factually true. It was just that the activity had dropped enough. The increased activity looked like a really sharp growth, but it was just kind of normalizing. Oh, okay, okay, interesting. Well, we are certainly hoping for smooth sailing in the Taiwan Strait. Let me tell you about figma. No matter where your idea starts, figma make clog code codex or sketch. The Figma canvas is where ideas connect and products take shape. Build in the right direction with figma. Asml. Bern Hobart Funny post here. ASML can't figure out how to make money from EUV machines, so they sell them to tsmc. But TSMC can't figure out how to make money from chips, so they sell them to Apple. Apple can't figure out a profitable way to use iPhones, so they sell them and there you go, the profit. And anyways, Dr. Kareem Carr, is someone saying Bear posting. Yes, Bear posting. That they don't know how to make money from AI directly. This is really. This is such a funny criticism. It's such a funny criticism because if they actually were going with this, the criticism would be insane. Insane. It would be like they created super intelligence and they're keeping it to themselves. Yeah, exactly. The whole point is that every single person on Earth, whether you pay for a plan or not, can benefit from today's models. Indeed. Well, let's head over to Meta. But first let me tell you about 11 labs. Build intelligent real time conversational agents. Reimagine human technology interaction with 11 labs. So Nebias and Meta have agreed to a $27 billion AI infrastructure pact deal. The talks are advanced to Pact stage. 5 year deal. 27 billion to supply AI infrastructure capacity to Meta. Nebius has really been an entire fascinating company. Formerly part of Yandex, spun out independent, now publicly traded, and just one of the NEO clouds that's figured out that Microsoft deal and now seems to be doing good work with Meta. So Navia said it will provide $12 billion of dedicated capacity across multiple locations. Meta will also purchase up to 15 billion in additional capacity over the five year. Over the five year period. These deals are sort of squishy, but it doesn't matter because the people who actually need to know can underwrite them. Accordingly, Nebius added that it will use large scale deployments of Nvidia's next generation Vera Rubin AI infrastructure, which Jensen is surely talking about at GTC right now. That's expected to be available in the second half of the year. And Nebias will begin delivery of that capacity beginning early next year, which feels like a decade in AI timelines. Why do you have the paper in front of your face? The team earlier said I look like a third base coach, so I'm covering up. Yeah, because you don't want to let everyone know what play you're calling. There you go. Exactly. There was news Friday, late a rumor or some reporting from Reuters, Metta is planning sweeping layoffs that could affect 20% or more of the company, three sources familiar with the matter told Reuters. As Meta seeks to offset AI infrastructure bets and prepare for greater efficiency brought by AI assisted workers. How many employees does Meta have? I think it's like 60,000, something like that. Let's figure. 75. 75,000. June 30, 2025. That's the same number. Yeah. 78,000 as of December 31st. Somewhere in the same range as Salesforce. And again, not super surprising. Stock's up around 2% today. I would expect this to pop even harder once these layoffs are actually announced. Yeah, I mean, the advice is become aligned with the AI effort at Meta. If these layoffs happen, they're clearly cutting part of the workforce, but then they're also acquiring and hiring all over the place. Just more around AI. I mean, we saw that today with the Manus announcement. They're taking new naming Meta. Just call your product a computer. Manus Computer, we got. No, no, no. It's called My Computer Manus. My Computer by Manus. My Computer by Manus. It works on mobile, works on your computer Manus Desktop. But wait again, this was My computer is the core feature of the new Manus desktop app. It's your AI API. Okay, so still called Manus direct competitor to Codex Claude Code Cowork and Microsoft Cowork. At this point, everyone's doing cowork, so maybe you just rip that. Yeah. So the reason I thought the Manus acquisition was interesting at the time is people were positioning it as more of a talent acquisition. Like, these are great product builders that figured out how to grow products super quickly. I think at the time they sold, they were somewhere in the range of 100 to 200 million of run rate. I was interested in it specifically because it seemed like Zuck was trying to take what they had built and actually just scale it, not just roll them into working on ads or whatever other products. Tyler, please download My Computer by Manus and play around with it and come back with a review. So the top recommended action that they showcase here is organizing thousands of unsorted photos. I'm not super into organization for the sake of organization, but that does seem pretty useful. I was taking photos on an actual camera this weekend and had to transfer them from the camera to an iPad, then scroll through them, favorite them, then share them over airdrop. And there is a cool agentic workflow, which is basically actually download the raws. Some of them were a little bit overexposed. Some of them need a little bit of color grading. And if I could have a workflow where Manus or some desktop agent opens every photo individually in Photoshop and tweaks it and does it intelligently and crops it ever so slightly and is thoughtful about it, that would definitely speed up my life. Dodd says would trust OpenGL more than Manus after the Meta acquisition with private data. Yeah, well, the Manus branding, the meta branding on this is so limited. I would be surprised if people sort of, you know, if this, if this goes broad, people wouldn't necessarily know that much. I wonder if they'll do the Oculus thing and you'll have to, like, log in with Facebook at some point. You can log into Facebook. You can already, but you can also log in with normal, like email. Yeah. I mean, man asked before was, wasn't it a Chinese company? It was based in Singapore, but it was, you know, like, like rumored to be aligned with China. And so, I mean, not rumored. They. They were building it in China. Okay. To Singapore, because the optics were not good. So, you know, as far as. As far as private data security goes, I think this is an upgrade. Right. It certainly feels like it. Anyway, let me tell you about Phantom cash. Fund your wallet without exchanges or middlemen and spend with the Phantom card. So the Oscars happened last night, Jordi, just to get you up to speed, the Oscars are an award show that are put on by the Motion Picture and Academy of Arts and Sciences. Yeah, I saw someone on the ramp cap table got an award. Yeah, yeah, yeah. Michael B. Jordan won best actor and he won best investor for that. Best Investor, Yeah. They should have a category for that. But Timothee Chalamet is getting taken to task in the Financial Times over his views on opera and ballet, of all things. The Financial Times writes, it's quite sweet, really. So desperate are some people to get their knickers in a twist on the Internet that in the face of a lull in the culture wars, we have real wars now. The only thing they have found to get outraged about recently relates to a man saying, nobody cares about ballet and opera anymore. The man I refer to as Timothee Chalamet, a talented young actor who stars in the multi Oscar nominated Marty supreme, which had a very unfortunate showing at the Oscars. I think they were nominated for nine awards and they didn't win anything. And so upset is your belief that it had to do with his comments disrespecting? No. Or it was just the people, the. The critics actually just said, hey, like, you know, Yeah, I think in every category he was. Marty supreme was up against like a Goliath. Like it was a. Every fight was sort of a David and Goliath and there were just no upsets because he was going up against sinners in one battle after another, which were heavy favorites, I think, from the very beginning before these comments were made. So Timothee Chalamet was talking with a fellow actor, Matthew McConaughey. At a town hall event organized by CNN and Variety in February. But the comments actually just got clipped and went viral recently. It was two week delay. The slicers over there got to step it up. He said. I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive. Even though, like, no one cares about this anymore. All respect to the ballet and opera people out there. And then he said distinctly, disrespectfully, I just lost 14 cents in viewership. Damn. I just took shots for no reason. There is evidence of Chalamet showing having made similar comments before, such as on the Graham Norton show in 2019, when he called opera a, quote, outdated art form. And at an event the same year where he was worried that cinema would become like opera or ballet or something, kind of a dying art form or something. He also, as many people, as many of those who claim to feel so offended have pointed out, has close family connections to the world of classical dance. His mother, grandmother and sister all danced with the New York City Ballet. Wow. And he has spoken out about growing up dreaming big backstage at the Koch Theater in New York, where the ballet performs. As someone who tried to pursue a career in pop music, while my older sister. This is the writer in the Financial Times, my older sister pursued one in classical piano. I would wager that he has been honing this particular attack, or perhaps defense line since adolescence. So his apparent instant regret, his slip, felt. Felt a bit disingenuous. Are you a. Are you an opera fan? Ballet fan? I like the opera. Me too. Although I actually haven't not been to opera yet, so it's hard for me to. And I just think, like, there's a world where, you know, the film and movie industry, like, does become like opera and ballet, but that's still like a beautiful thing with an amazing culture. We've seen this in la, where I think a lot of movies are now releasing only at these, like, kind of fancy theaters. Yeah, Tarantino has one. Or the Chinese theater. These kind of things where it's like much more like kind of upstage and a real event that you go to. Yeah. And of course it is. It is like, you know, just technological disruption with social media and there's a lot of other, like, gyrations in the transition there. But I'll tell you why I think this whole kerfuffle's happened. Kerfuffle happened. And as someone who doesn't really follow Hollywood, doesn't follow film, doesn't follow Timothee Chalamet et cetera, et cetera. I think what is happening is he came out with this new it's okay to pursue greatness on the path to greatness. Sure, sure, sure. I'm trying to be the goat. I'm trying to. With this kind of like, bravado. Yeah. And if you do that and it's like, me, me, me, me, me, me. I'm trying to be the greatest. And then you start just randomly taking shots at another art form where other people are pursuing greatness. Sure. You just invite a lot of criticism because I think, like, everyone's okay, I think with somebody like, you know, being on their own personal pursuit of greatness. But if you're doing that while trying to tear down other art forms. Yeah. You're just gonna invite massive criticism. Yeah. It does feel like he's sort of. He's sort of collapsing, like market cap and like tam of like. Yes. The opera tam and the ballet tam is smaller than film, but it would be odd. Play the actual sound. Yeah. Let's play for people that are here that are younger than me, were people desire, are desiring things that are more patient and that pull you in. I just saw another article that says Gen Z is a bigger movie going audience than a millennial audience. You know, I feel like a fucking grandpa saying that. No, but point being, I think even like Frankenstein, which is like a hugely popular movie this year, I didn't think that pacing was extraordinarily fast or anything, but it pulled people in, you know. But it does take you having to wave a flag of, hey, this is a serious movie or something, and some people want to be entertaining quickly. I'm really right in the middle, Matthew, because I admire people and I've done it myself to go on a talk show. Hey, we got to keep movie theaters alive. We got to keep this genre alive. And another part of me feels like if people want to see it, like Barbie, like Oppenheimer, they're going to go see it and go out of their way to be loud and proud about it. And I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive. Even though no one cares about this anymore. All respect to the ballet and opera people out there. I just lost 14 cents in viewership. But crazy shots, that's not a shot. I hear what you're saying. Yeah, Yeah, yeah. I don't know, it's interesting. I was thinking about if the creator of GTA 5, like, stood on stage and was just like, we are 10 times the size of the baseball. Baseball. But also, like the movie industry, like the gate. The video gaming industry has been basically 10 times the size of the. Of the movie industry for. You mean the movie theater business? No, like. Like Hollywood. Like gross production. Yeah, totally. I'm almost positive. Not. Not 10 times the size of your streaming platform. Yeah, maybe streaming. That includes TV shows. And then do you include mobile games or not? That's a big question. But the video game industry is definitely bigger. Raghav in the Twitch chat From deep Nvidia CEO just said he sees 1 trillion in revenue through 2020. Bring down the gong. Bring down the mallet. Congratulations. That's amazing. Thank you. And we have our next guest in the Restream waiting room. First, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And we are joined by Kevin from Epic Gardening in the Restream waiting room. Let's bring him in the TV show. Kevin, how are you doing? What's going on? What's up, brothers? How you doing? Good to see you, brother. Thanks so much for taking the time to join the show. First up, we gotta talk about that tank. Yeah. What's in the tank? What's in the tank we've been talking about? I'm breeding rare Costa Rican tree frogs in this tank. No way. Yeah, they're endangered, okay. Oh, they're endangered. Okay. What else is. What else is special about. And you're planning to release them in all 50 states. States. Once you have enough. This is the goal. This is the goal we're always scaling over here. Yeah, of course. I love it. I love it. Is it challenging? Like, how much of your time is devoted to that particular tank? Almost none. Almost none. I just need to make sure that they're fed. Yeah, that's cool. What do they eat? Yeah, they eat crickets, which I'm breeding in that little tank right over there. You can see that. Yeah, we're breeding the different trophic levels over here for sure. Okay, and then. And then. Do these frogs have use in your garden? Is it purely just for fun? No, I'm just branching out to flora or to fauna now, I guess. Yeah. Here at Epic, you know. Okay, well, first time in the show, so I want to kick off with your backstory. I want to know about the decision to start making content. I feel like that's always an interesting origin story. Like, when did you think. Okay, I Need to make content, dude. I mean, I'm an Internet OG, so I was on GeoCities, I was on Angelfire back in the day, anime tutorials, you know, so I don't know what it is. I think genetically I'm designed to make content, but for Epic, it was really a calling card for. Remember when you used to design WordPress websites back in the day? Like when people actually paid for that service? I used the blog as like a calling card or a digital business card for, like, designing websites for local businesses and then just kind of kept plugging along with it and adding different platforms and here we are today. Yeah. What about the first YouTube video? Like, what was the backstory behind choosing to go to YouTube? Choosing to go to video. It's a big lift for people if they're on substack or they're a writer and they don't know how they're going to do in front of camera. First YouTube video was 2013, so it was a long time ago. And ironically, back then, I mean, SEO and blogs were kind of the thing. And so for me, the first YouTube video, maybe it's the second YouTube video. You can see me using a screen recording app, reading a blog article, just literally reading the article and with the hopes that people would watch that video and click the blog link and I would make money off of the advertising on the blog. So it's a completely backwards logic to today, of course. Yeah, yeah, yeah. But then obviously discover YouTube is a far better platform, especially these days.

Growth. But of course, you know, risks to all of this. One dozen over on X says they were right to take cigarette ads off tv. I would have smoked a pack a day if I saw this when I was 14. What is this video? Pull it up. Is this real ad? This cannot be a real ad. I think it gets a sound, some sort of vibe. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that actually, I'm so confused by this ad. I think that's Charlie Sheen, right? Is that the arc de trio paris? Good music, though. This should be the new launch video. Meta. Oh, it was an international ad. The message there is that they're taking New York to France. But then it was Japanese text on screen. I don't know, it seems like some sort of mashup.

Spell checking names and cross referencing data sources and pulling from email and dumping some notes, summarizing. And you can do that on a GPT4 class model instead of using 5 4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and so that's still economically valuable and so you're able to continue that. Should we go over Ben Thompson's post from this morning? Yeah, we should. Now be a good time. Yeah. First, let me tell you about Label Box, RL environments, Voice robotics, evals and expert human data. Label Box is the data factory behind the world's leading AI teams. And let me tell you about Vibe Co, where DTC brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audiences, measure sales. Just like on Meta Agents published this this morning. To me, the second I saw that I started reading it. It felt like taking a double scooter with C4. Is that a pre workout? Yeah. You never. I know the can. I didn't know it was a. You never, you never dabbled. What was the one that we. I'm more of the gorilla mind one. That's the one that I do. Many people have said you have the mind of a gorilla. Yes, yes, yes. For more plates, more dates. You're a gorilla in sheep's clothing. I think that's literally the pre workout that I have, although I don't use it that often anyway. So you got pumped up? I got pumped up. Ben writes, there's a weird paradox in terms of AI prognosticization. Prognostication. Prognostication. That was a good, good effort, Jordy. On one hand, there's some. There's just so many words requirements for having a podcast, like knowing how to say words. No taste. I mean, yeah, ultimately there's a lot of words that you, when you read them. Yeah. You're just like, oh yeah, you can just do it. And then you try to rip it. On one hand, you don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out to be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I very, I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026. However, on the morning of Nvidia GTC I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gun going. He writes LM paradigms over the last couple of weeks, first in the context of Nvidia earnings, and then last week in the context of Oracles. I've talked to you about 3 LM inflection. I've talked about 3 LM inflection points. Yeah, I'm not going to go through all of these. He goes chat. We've talked about this a few times. Reasoning models and then agents. And each one of those increases the demand exponentially for computer. So chatgpt01 and then opus as well as Claude Code and Codex. Codex basically getting to the point where tasks are being accomplished over hours and getting to great outcomes. And this is the interesting point, the Decreased need for Agency the reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens, because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the models so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of capex expenditure not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that computer and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they are going to be used even more than even reasoning models in a chatbot. It's how this third point will be manifested that I think is underappreciated. After all, far more people use chatbots than agents, but I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to Use AI. And he goes into a little bit talking about local compute, talking about how Apple's opportunity to run LLMs locally. There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599, I think $499 for education. Potentially very disruptive to other laptop makers. You said you still get discounts, Tyler, or does it. I think I'm still stabbed. Oh, yeah, you're still. I'm on leave because you're on leave. That's great. There you go. There are some legendary leave of absences where people have been away for like 10 years and then they go and do so many. See, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before, while you're still eligible. That's a good one. Because they're like, oh, well, we got it. I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year, and I think they gave him the honorary degree a couple years later. So, you know, that's the. That's the speedrun to beat. But the point about the MacBook Neo is that at 599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with, like, Asus or another brand? I'm not in the Apple category. Like, it's not an option. Because that store over there, those. Those laptops start over 1,000. That's not my budget, so I'm not even going in that store. Well, now you can. And you can spend $600 and get a pretty good computer. And the CFO Nick Wu of Asus was on their recent ear and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments in the second half of last year. We made some internal prep, but now that it's out, we don't think it's that big of a deal. It has some limitations. Specifically, it only has eight gigs of ram. So, like, you know, this is more focused on content consumption. It's not a mainstream notebook for notebook usage, for creation, for working. It's not a workbook device. It's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, that's what people use These laptops, for now, it is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more. And so. And then he goes on to apply that to AI, talking about enterprise and the value of. Companies have a demonstrated willingness to pay for software that makes their employees more productive. And AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, basically my interpretation, he's making a case that there are companies that could cut headcount and actually just grow faster. Yeah, if. If they're implementing AI properly, not just replacing, like the routine workloads. Yep, so he says. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay, actually, I'm going to start one paragraph. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through a huge apparatus filled with humans who accelerate the effort in some vectors and retard it in others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard to manage and motivate human cogs in the organizational machine with agents that not only do what they are told, but do so tirelessly and continuously until the job is done. This only makes the argument that we are not in a bubble much more compelling. First, unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously. I'll. I'll get around to it when I feel like it. I'll give it a crack. I'll get around. Yeah, this only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of LLMs are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing. Right. You have one. Tyler just you know, he's going to. Tyler's going to set up to be able to do sign language with his agents to just. Just be. Not even speaking, just. Just send. Do you actually ever use any of the voice models? Remember Carpathy was talking about that? Yeah, A lot of people do this because you can just like talk much faster, I guess. I haven't done this really. I've used it sometimes use the voice mode, but I don't have to use it in like coding agents yet. I was using the ChatGPT voice mode, like the True, like back and fourth voice mode. Real time voice. Real time voice mode in the car this morning. And they improved that thing dramatically. It's good. Yeah, it's so much better. So first off, it doesn't do that. Like that's a great question or anything like that. Or that whole pause that was in the super bowl ad. Like that just doesn't exist anymore. It just answers and it answers in these like really short, punchy things. I was asking it about like, how many jobs are actually in America? And it just says like 164 million. And it just like gets me the answer. And I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more detailed back and forth without needing to like dictate a whole prompt and then let pro cook on it for 10 minutes, come back, have it read it. To me, it was like a much better experience. I was very pleasantly surprised by how the back and forth worked. And they also changed it so that you see the floating bubble of like the little animation, but the text populates in real time with the. With your question and then the answer and then your question and the answer. So you can just scroll and read as well. It's very cool. Anyway, third, the last argument that we are not in a bubble. That economic returns from using agents aren't just impactful on the bottom line, that is saving on cost, but the top line as well. Let's go in this context. It is any wonder that every single hyperscaler says the demand for compute exceeds supply and that every single hyperscaler is in the face of stock market skepticism, announcing capex plans that blow away expectations. So I encourage you to go subscribe to Strategy, max out your plan, pay annual. But this was extremely notable. It's such a funny ending where he has this point about like, you only need to be worried about a bubble when you don't need to be worried about a bubble. If everyone's saying a bubble. Because then everyone's like, risk off. Because everyone agrees that, oh, we're in a bubble, let's not do bubble behavior. And so capitulation is the sign of a bubble. And he's like, I understand that. And still this is my take. It's a bold take, but I think it's a good one. Really quickly, let me tell you about the New York Stock Exchange. Want to change the world, Raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets. Take us through it, Jordan. He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets. You always want to be, he says. I literally think all the marks are wrong. Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives of the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients. Late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary. According to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public. I don't know. Also detailed calling, calling, you know, people in private markets arrogant. It's crazy. I feel like, I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold, bold call by him. But we'll see what he, what evidence he has to back up that extraordinary claim. He blamed the media for creating a frenzy private credit. Obviously we're in the middle of a private credit party. Apparently if you do credit well, it's honestly, I would say we don't understand private credit well enough to like really put, put everyone up into a frenzy. He says if you do credit well it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. Zito talked about the sell off in shares of large software companies which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity, many with private credit loans, could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good, so all is good. Anyone who says that clearly doesn't understand. Most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public yet, smaller than those companies and we're trading at a much higher valuation than those companies. And so I am concerned about many of those. Take privates. Yeah, I remember a lot of like Logan Bartlett was doing a ton of analysis in the end of the ZURB era at how high the multiples were in the public markets and that's what was driving the 100XR transactions. And you have to imagine that even, even if we, we were like oh yeah, that, that you know, VC backed company was sort of over, overhyped at 100x RR. Well that still has a trickle down effect to you know, the private equity buyout. That's just like yeah. Remember last year when Figma went out? Yeah. And they priced it. Yeah. Very reasonably. Right. They were very intelligent how they priced it. But then obviously there was so much excitement. It's such a great company. Ran up when it. In the first couple days there was, there were some late stage private SaaS companies that I remember were posting like API to go public. Yeah. I think it was the Parker rippling was like oh if I can get. That's a crazy multiple. Yeah, if I can get some insane revenue multiple maybe this appealing. Obviously a lot of those names still could get out this year. Strong companies. They're not, not as, as eager to get out. Zito pointed To Thoma Bravo's 2021 6.4 billion take private of the software firm Adalia. In particular, several lenders to Medallia, including Apollo, have already written down its debt. He says there will be an issue with respect to that credit which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic Small or mid sized. Joe Software company Zito said. Joe Software company. If he's in the wrong place, I think he's going to recover somewhere between 20 and 40 cents. So 60 to 80% markdown. A lot of the private credit firms have been they'll mark down alone but like mark it down to like 95. Sure you know, sure you know nothing very significant. Zito noted that he expects private credit loans originated in the next 12 to 18 months to be a much better vintage as it relates to quality of company, amount of leverage, documentation and spread. He also weighed in on redemptions and whether private credit managers should enforce limits, typically 5% of a fund's shares each quarter or allow more investors to cash out when they are flooded with requests. It is a topic he and others on Wall street have recently been asked about. As funds take different approaches, you're going to see elevated redemptions for a handful of quarters. I don't know how long it lasts. Making a decision in one quarter may be the right decision for fundraising in the near term and then a quarter later you'll realize it was a really bad decision. So my overall bias is to stick to the 5% to protect all of my existing investors on vulnerabilities in private equity. Zito sought to shift the focus to private equity, where Apollo has less exposure than most of its peers. He suggested investors voracious demand for buying stakes in existing private equity investments. But wariness of the private debt underpinning those deals doesn't add up since the equity would be junior to the debt if there were major problems with these assets. There's unlimited demand for secondary private equity, but they're worried about private credit, which finances 80% of those portfolios. I can't compute, but I'm the dumb guy. I don't understand. I start saying this and I get these blank stares back at me like, okay, I don't know. He said, I literally think all the marks are wrong. Is that what you're asking me? I think private equity marks are wrong. And again, I read into this. He's talking so candidly, at least in private equity land, he doesn't feel exposed enough to be freaking out. Yeah, a couple more quotes. He says this next cycle is going to be a big moment in time for the private markets because people are way smarter than I think private market participants, particularly people in the wealth channel. Like, I kind of sense an arrogance of the people who grew up in the private markets business. If you don't mark your book, I think you actually lose trust with the clients. We're going to be the market leader in actually marking our book. Let's give it up for being the market leader on the economy and markets. He said. I think it's more likely than not that we go into a recession, a consumer confidence led recession. Most of the companies we lend to are getting a lot of pressure to show clear AI execution. It's forcing people to do stuff before the actual technology works. That's going to be the first step of just a slowdown in the broader economy. Interesting. He said. I literally think Powell, he's so upset at how it's ending that he's just saying there's inflation every day to piss off the president. Like I literally think that's what's going on. It's so hard for me to see inflation. I don't see it anywhere. I see it much more deflationary. I think the technology is attacking every profit pool. What do you say? Asked why a popular high yield corporate bond ETF seen as a benchmark for such debt that is typically under pressure in an economic crunch was relatively flat for the year says I don't have any idea. The amount of dispersion going on beneath the surface is kind of crazy. I literally at home I told my wife last night I feel like the market should be down at least 10% and it's flat or up. Can't make heads or tails of it. On Apollo's credit business he says on our balance sheet we are 95% investment grade, private and public investment grade. I have a view that bigger companies are going to do better than smaller companies. And so I've tried to position my. The terminology gets me every time. Yeah, I know because is there like you asked me, I run a private credit fund. We mostly back. We mostly invest in non investment grade opportunities. Yeah it's like brother, don't you want to be. What were you doing? It's in the name now of course. Very very end of this journal is they have a form. We want to hear from you. Are you currently an investor in private credit funds or are you planning to become one? We'd like to hear from you, share your thoughts or experiences in the form below. They're looking for snitches. Yeah. I'm going gig along back to data center land. Let me tell everyone about console consul builds AI agents that automate 70% of it HR and finance support giving employees instant resolution for access requests and password resets. And let me also tell you about CrowdStrike. Your business is AI. Their business is securing it CrowdStrike secures AI and stops breaches. Got to give another shout out to George Kurtz, who went 1 and 2 again at the Chinese GP over the weekend. Mercedes on an absolute tear. It's the Kurtz effect. I saw a. I saw some sort of promotional post for a vintage Le mans Racing Series. So 24 hours, but there's some date where, like all the cars have to be from early, before 1990 or something like that. I don't. I don't exactly know how old. I didn't dig into it, but it looked very, very co. Anyway, Cerebrus just landed aws. Amazon announced inference chips deal with Cerebras, which is big. They are proving the doubters wrong. Elon is saying that the terrafab project launches in seven days. Bev Jesus says what? Very, very fast timeline. Obviously, when people heard about his plans on Dwarkash, a lot of people kind of questioned it, but Elon's used to being questioned. Yeah, Cerebras is such a cool company. Like just the first time. I mean, we've seen it with like the chat, Jimmy AI and, you know, just going to Codex Desktop, which is of course like a coding harness, but you can just ask it questions and you can experience Codex.5. I think 5.3 is on. Yeah, 5.3. Spark. Spark. Yeah, Spark. And it just gives you the answer immediately. And it's actually very, very magical. And I think that's going to be really good for retention. Basically everyone's going to be in the smiling curve. Will smile more as people come back. Matthew Zeitlin says data center capacity growth is slowing. He's pulling data here that says newly added US data center capacity slows down considerably in Q4 20, 25, as market struggles to keep up with explosive demand. 25 gigawatts of data center capacity added to the funnel in Q4, 50% less in Q3. We'll see if that ramps up again or. That still seems like a lot like the number that I was hearing was for this year. The target for anthropic is like 5 gigawatts, which is like an insane amount of compute, but at the same time, like in the context of 25 gigawatts in one quarter, like, it feels like, like the. There is like still significant growth, but of course, you know, risks to all of this. One dozen over on X says they were right to take cigarette ads off tv. I would have smoked a pack a day if I saw this when I was 14. What is this video? Pull it up. Is this real ad. This cannot be a real ad. I think it's some sort of vibe. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that actually, I'm so confused by this ad. I think that's Charlie Sheen. Right? Is that the arc de trio, paris? Good music, though. This should be the new launch video. Meta. Oh, it was an international ad. The message there is that they're taking New York to France. But then it was Japanese text on screen. I don't know. It seems like some sort of. Some sort of mashup. I don't know. Let's go over to Tyler Cowen. Yes. With how to lose the AI he wants you to lock in. Oh, no, I was going to go to his. Why you should work much harder right now. Okay. Over on Marginal Revolution before we get into the next piece. He says if strong AI will lower the value of your human capital, your current wage is relatively high compared to your future wage. That is an argument for working harder now at least if your current and pending pay can rise with greater effort. Not true for all jobs. If strong AI can at least potentially boost the value of your human capital, you should be investing in learning AI skills right now. No need to fall behind on something so important. You also might have the chance to use that money and buy into the property, into the proper capital and land assets. So work harder. He should have put this into a course. I would. I would follow this advice if it cost me $999 in six installments. Yeah, but because he's giving away it for free, it can't. Can't actually be that valuable. Kidding. Of course. From Ricardo in the comments. Suppose you are the best maker of horse carriages in Belgium. Around the time the automobile is invented, you might want to take on as many orders as possible for new carriages because, you know, future is precarious. Or maybe you get your hands on one of these newfangled automobiles as soon as possible and learn how to fix them. Both options require you to work harder, but these seem to be the two best options available. Paradoxical but true. That's a good take. I like that. A little bit of a white pill. Never, never a bad idea to work harder. Never a bad idea. Should we go through this? Yes, we should. First, let me tell you about Cisco. Critical infrastructure for the AI era. Unlocks seamless real time experiences and new value with cis. And let me also tell you about cognition. They are the makers of the AI software. Engineer Devon. Crush your backlog with Your personal AI engineering team. Where do you want to go next, Jordy? How to lose the AI arms race. Let's do it. So investor Leopold Ascherbrenner is now famous for situational awareness. His essay predicted that major AI companies would end up functionally as part of the government led national security project, possibly even nationalized. Along related lines. Nick Economist Noah Smith recently asked a critical question. If AI is a weapon, why don't we regulate it like one? We already know this is Tyler writing in the Free Press, the fp.com we already know that the Pentagon has been using Anthropic's Claude to interpret collected intelligence data and help plan the attacks in Iran. Advanced AI can also be used for cyber attacks, enemy surveillance and identification, followed by missile or drone attacks. Under most extreme scenarios, which may or may not be realistic, advanced AIs might design bioweapons, disable the nuclear weapons of an AD adversary by disrupting chains of command, or perhaps design and build a scheme to knock missiles out of the sky. Washington D.C. is starting to ask very basic questions about where we are, what we are doing here. Anthropic and the Department of Defense are at loggerheads over whether Anthropic's AI should be banned from government work. Senator Bernie Sanders recently raised a broader set of concerns calling for a moratorium on AI data centers with the intention of slowing down progress in artificial intelligence models. But circa 2026, neither nationalization nor an AI slowdown are feasible strategies for the United States. We need to keep our lead both in military and civilian uses of the tech. And that requires a dynamic private sector building our artificial intelligence models. Our federal government, working through the Manhattan Project, developed and built the first atomic bomb. But the strongest AI models are creatures of the private sector, whether we like that fact or not. Even China, which is far more statist than the United States, has seen its cutting edge models built by companies, not by the government. The top AI models are far too complex and require too much high paid talent, including, including international talent, to be done well by governments. Governments sometimes can succeed in building out massive hardware projects with the space program being another example. They are very but there are very few cases of governments succeeding with advanced software on a large scale. For that you need private sector dominance. There is no easy way to switch from that mode of organization, which includes salaries of tens of millions of dollars for top researchers, to a more bureaucratic approach. An attempt to do so would destroy or take down those companies, thus thwarting our standing in this new arms race. The general Reality is this. We all benefit from living in an advanced civilization rather than eking out subsistence as our ancestors did. But there is part of this bargain we have tended to ignore or take for granted. Now that human beings have developed advanced technologies, we, the freer and better societies, must commit to keeping the technological lead. We have not stayed ahead in every area of tech, but we need to be able to protect ourselves and our allies. It's a good thing that America built an atomic bomb before either Hitler or Stalin did. To the extent you believe AI is important, is important for weaponry and national security. That means we need to keep up the pace of progress. You might find that a slightly. You might find that a slightly unpleasant thing thought, because even under positive visions of an AI future, it will change our world a good deal. Nevertheless, it is a part of technology, of a technological bargain we have been living with for a long time. Arguably since the widespread deployment of firearms or explosives, we seem to have been lulled into a state of stupor by the longstanding technological dominance of the United States after World War II. In essence, we have to fight and win yet another arms race. You can't blame AI for that. You can blame AI for real for that reality if you want. But the reemergence of competitive arms races was inevitable with or without AI. You should redirect your ire toward modern history itself. AI might, may have accelerated the world's new arms race, but there are many other technologies that could play and yet may yet play a comparable role. Space weapons, anyone? How about lasers? Are new types of hypersonic missiles. At least with AI, the US currently holds the lead. The creativity behind top AI models plays into our national strengths. And he closes by saying so today we need an odd complex, an odd and complex mix of not entirely consistent ideologies for the current arms race to go well. How about some tech accelerationism mixed with capitalism and then a prudent technocratic approach to military procurement to make sure those advances serve national security ends. On the precautionary side, we need a dash of 1960s and 70s new left and libertarian anti war ideologies skeptical of Uncle Sam himself. We do not want to become the bad guys. Do you think we can pull that off? The new American challenge is underway. Inspiring. I like this. There was a lot of back and forth around the anthropic Department of War debate and Door Cash had a great piece on it and lots of people have chimed in now that like the dust has settled a little bit. And I think this is a good sort of nuanced take it doesn't boil itself down to a tweet just yet, but I think we are getting somewhere with the different trade offs that are at stake. What do you think, Tyler? Yeah, this is good. I mean, I think the whole thing that I basically got to when I wrote like the nationalization thing was that like there's just this. There's pretty big scale, right, of like what actually means to nationalize something. There's like the Manhattan Project, which is like, okay, this is like full scale, top down. Everything is decided by it, by one person. It goes down the pyramid and then there's like the very, you know, kind of distributed like, oh, like intel. Is that nationalization? Yeah, I think I broadly agree. Like, I don't think really, I don't think the Manhattan Project is really the best way to do this. Right. Because if you take like, you know, Tata Council thing is like, you know, state capacity libertarianism, like, is the government fully capable of continuing this AI progress that we have right now? Would us stay in the lead if the whole thing is set by the government? It's unclear. This is sort of what I was going back and forth with Karp on was people have framed this as a battle between Dario Amadei and Pete Hegseth. And I feel like we are a democracy. And so like, I would like more, more authority to be assigned to the individual American voter for a lot of these things. You know, you have that joke about like, we got to talk about it, we got to just talk about this. Like, what are we going to do about AI? It's like, well, like, we can, we can actually vote on it. Like, you can, you can have a plan and then people can vote for it. And there are a bunch of different ways to exercise political will. And it feels like there, there is a trade off, but we got to a good place with the nuclear weapons one. And I do feel like I, as a voter, I have a very small stake, one of 300,000. You know, I guess I have 300 million. I guess there's like 160 million people that vote in the national election. But part of the national election is, you know, do you trust this particular person to have the nuclear football, to have their finger. They're going to have their finger on the button. Like, well, let that sit with you before you cast your ballot. And it will be a continuation of that, like this, this is the person that will decide AI policy. So vote according to that. Right. And I hope that there's more, more of a, of a understanding that the American voter, the American citizen does have a huge stake in the AI future. And it's not just the, the like, you know, the, the high flying personalities that give, you know, speeches and podcast appearances. There is a lot more to the American project than that. Well, there is a lot of news around Taiwan and what might happen over there. We found an interesting kal she market that sort of tracks just general unrest in Taiwan. So the question is, will the United States issue a level four travel advisory for Taiwan? That of course would be a very, very bad news if that did happen. It's sitting at 46% before 2028, January, January 1, 51% before 2029, and 57% above for 2030. And so this is sort of a way to understand geopolitical risk. Obviously, we hope this calms down and this market goes to zero. Because you sent me that, you sent me that headline about increased activity around Taiwan. Yeah, some of it was fake news. Yeah, some of it. I think the reason that it triggered really calling me out here, sending you fake news. You're like you actually fell for a viral hoax recently. No, I mean, I looked at it and it was factually true. It was just that the activity had dropped enough that the increased activity looked like a really sharp growth, but it was just kind of normalizing. Oh, okay. Okay. Interesting. Well, we are certainly hoping for smooth sailing in the Taiwan Strait. Let me tell you about figma. No matter where your idea starts, figma make cloud code, codex or sketch. The figma canvas is where ideas connect and products take shape. Build in the right direction with figma. Asml. Bern Hobart, Funny post here. ASML can't figure out how to make money from EUV machines, so they sell them to tsmc. But TSMC can't figure out how to make money from chips, so they sell them to Apple. Apple can't figure out a profitable way to use iPhones, so they sell them and there you go, the profit. And anyways, Dr. Kareem Khar, is someone saying bear posting. Yes, Bear posting. That they don't know how to make money from AI directly. This is really. This is such a funny criticism. It's such a funny criticism because if they actually were going with this, the criticism would be in insane. It would be like they created super intelligence and they're keeping it themselves. Yeah, exactly. The whole point is that every single person on earth, whether you pay for a plan or not, can benefit from today's model. Indeed. Well, let's head over to Meta. But first let me tell you about 11 labs build intelligent real time conversational agents. Reimagine human technology interaction with 11 labs. So Nebias and Meta have agreed to a $27 billion AI infrastructure pact deal. The talks are advanced to Pact stage five year deal. 27 billion to supply AI infrastructure capacity to Meta. Nebius has really been on terra, fascinating company. Formerly part of Yandex, spun out independent, now publicly traded and just one of the Neo clouds that's figured out that Microsoft deal and now seems to be doing good work with Meta. So Nabia said it will provide $12 billion of dedicated capacity across multiple locations. Meta will also purchase up to 15 billion in additional capacity over the five year. Over the five year period, these deals are sort of squishy, but it doesn't matter because the people who actually need to know can underwrite them accordingly. Nebius added that it will use large scale deployments of Nvidia's next generation Vera Rubin AI infrastructure, which Jensen is surely talking about at GTC right now. That's expected to be available in the second half of the year. And Nebias will begin delivery of that capacity beginning early next year, which feels like a decade in AI timelines. Why do you have the paper in front of your face? The team earlier said I look like a third base coach, so I'm covering up. Yeah, because you don't want to let everyone know what play you're calling. There you go. Exactly. There was news Friday, late a rumor or some reporting from Reuters. Metta is planning sweeping layoffs that could affect 20% or more of the company. Three sources familiar with the matter told Reuters. As Meta seeks to offset AI infrastructure bets and prepare for greater efficiency brought by AI assisted workers. How many employees does Meta have? I think it's like 60,000, something like that. Let's figure 75. 75,000. June 30, 2025. Yeah, 78,000 as of December 31. Somewhere in the same range as Salesforce. And again, not super surprising. Stock's up around 2% today. I would expect this to pop even harder once these layoffs are actually announced. Yeah, I mean, the advice is become aligned with the AI effort at Meta. If these layoffs happen, they're clearly cutting part of the workforce. But then they're also acquiring and hiring all over the place, just more around AI. I mean, we saw that today with the Manus announcement. They're taking new naming Meta. Just call your product a computer. We got Manus Computer, we got Perpetual Computer. It's called My Computer Manus My Computer. By Manus. My Computer by Manus. It works on mobile, works on your computer. Manis desktop. But wait again, this was my computer is the core feature of the new Manus desktop app. It's your AI API. Okay, so it's still called Manis. Direct competitor to Codex Claude Code Code Cowork, and Microsoft Cowork. At this point, everyone's doing cowork, so maybe you just rip that. Yeah. So the reason I thought the Manus acquisition was interesting at the time is people were positioning it as more of a talent acquisition. Like, these are great product builders that figured out how to grow products super quickly. I think at the time they sold, they were somewhere in the range of 100 to 200 million of run rate. I was interested in it specifically because it seemed like Zuck was trying to take what they had built and actually just scale it, not just roll them into working on ads or whatever other products. Tyler, please download My Computer by Manus and play around with it and come back with a review. So the top recommended action that they showcase here is organizing thousands of unsorted photos. I'm not super into organization for the sake of organization, but that does seem pretty useful. I was taking photos on an actual camera this weekend and had to transfer them from the camera to an iPad, then scroll through them, favorite them, then share them over airdrop. And there is a cool agentic workflow which is basically actually download the raws. Some of them were a little bit overexposed. Some of them need a little bit of color grading. And if I could have a workflow where Manus or some desktop agent opens every photo individually in Photoshop and tweaks it and does it intelligently and crops it ever so slightly and is thoughtful about it, that would definitely speed up my life. Dodd says would trust OpenGL more than Manus after the meta acquisition with private data. Yeah, well, the Manus branding, the meta branding on this is so limited. I would be surprised if people sort of, you know, if this. If this goes broad, people wouldn't necessarily know that much. I wonder if they'll do the Oculus thing and you'll have to, like, log in with Facebook at some point. You can log into Facebook. You can already, but you can also log in with normal, like, email. Yeah. I mean, Manus before was. Wasn't it a Chinese company? It was based in Singapore, but it was, you know, like. Like rumored to be aligned with China. And so, I mean, not rumored. They. They were building it in China. Okay, okay. To Singapore. Because the optics were not good. So, you know, as far as. As far as private data security goes, I think this is an upgrade, Right? It certainly feels like it. Anyway, let me tell you about Phantom cash. Fund your wallet without exchanges or middlemen and spend with the Phantom card. So the Oscars happened last night, Jordi, just to get you up to speed, the Oscars are an award show that are put on by the Motion Picture and Academy of Arts and Sciences. Yeah, I saw someone on the ramp cap table got an award. Yeah, yeah, yeah. Michael B. Jordan won best Actor and he won best Investor for that. Best Investor, yeah. They should have a category for that. But Timothee Chalamet is getting taken to task in the Financial Times over his views on opera and ballet, of all things. The Financial Times writes, it's quite sweet, really. So desperate are some people to get their knickers in a twist on the Internet that in the face of a lull in the culture wars, we have real wars now. The only thing they have found to get outraged about recently relates to a man saying, nobody cares about ballet and opera any more. The man I refer to as Timothy Chalamet, a talented young actor who stars in the multi Oscar nominated Marty supreme, which had a very unfortunate showing at the Oscars. I think they were nominated for nine awards and they didn't win anything. And so upset is your belief that it had to do with his comments disrespecting? No. Or it was just the people, the, the critics actually just said, hey, like, you know, yeah, I think in every category he was. Marty supreme was up against like a Goliath. Like it was a. Every fight was sort of a David and Goliath and there were just no upsets because he was going up against sinners and one battle after another, which were heavy favorites, I think from the very beginning before these comments were made. So Timothee Chalamet was talking with a fellow actor, Matthew McConaughey, at a town hall event organized by CNN and Variety in February. But the comments actually just got clipped and went viral recently. It was two week delay. The slicers over there got to step it up. He said, I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive. Even though, like, no one cares about this anymore. All respect to the ballet and opera people out there. And then he said distinctly disrespectfully, I just lost 14 cents in viewership. Damn. I just took shots for no reason. There is evidence of Chalamet showing having made similar comments before. Such as on the Graham Norton show in 2019, when he called opera a, quote, outdated art form. And at an event the same year where he was worried that cinema would become like opera or ballet or something, kind of a dying art form or something. He also, as many of those who claim to feel so offended have pointed out, has close family connections to the world of classical dance. His mother, grandmother, and sister all danced with the New York City Ballet. Wow. And he has spoken out about growing up dreaming big backstage at the Koch Theater in New York, where the ballet performs. As someone who tried to pursue a career in pop music, while my older sister. This is the writer in the Financial Times, my older sister pursued one in classical piano. I would wager that he has been honing this particular attack, or perhaps definitely fence line, since adolescence. So his apparent instant regret, his slip, felt. Felt a bit disingenuous. Are you a. Are you an opera fan? Ballet fan? I like the opera. Me too. Although I actually haven't not been to ballet opera yet, so it's hard for me to. And I just think, like, there's a world where, you know, the film and movie industry, like, does become, like, opera and ballet, but that's still like, a beautiful thing with an amazing. You've seen this in la, where I think a lot of movies are now releasing only at these, like, kind of fancy theaters. Yeah. Or the Chinese Theater. These kind of things where it's, like, much more like kind of upstage a real event that you go to. Yeah. And of course, it is. It is, like, you know, just technological disruption with social media. And there's a lot of other, like, gyrations in the transition there. But I'll tell you why I think this whole kerfuffles happen. Kerfuffle happened. And as someone who doesn't really follow Hollywood, doesn't follow film, doesn't follow Timothee, Chalamet, et cetera, et cetera. I think what is happening is he came out with, like, this new, like, it's okay to pursue greatness. Yeah. On the path to greatness. I'm trying to be the goat. I'm trying to, you know, like, coming out with this kind of, like, bravado. Yeah. And if you do that, and it's like, me, me, me, me, me, me. I'm trying to be the greatest. Yeah. And then you start just randomly taking shots at another art form where other people are pursuing greatness. Sure. You just invite a lot of criticism, because I think, like, everyone's okay. I think with somebody like, you know, being on Their own personal pursuit of greatness. But if you're doing that while trying to tear down other art forms. Yeah, you're just gonna invite massive criticism. Yeah, it does feel like he's sort of. He's sort of collapsing like market cap and like tam of like, yes, the opera tam and the ballet tam is smaller than film, but it would be odd. Play the actual sound. Yeah, let's play for people that are here, that are younger than me, were people desire, are desiring things that are more patient and that pull you in. I just saw another article that says Gen Z is a bigger movie going audience than a millennial audience. You know, I feel like a fucking grandpa saying that. No, but point being, I think even like Frankenstein, which is like a hugely popular movie this year, I didn't think that pacing was extraordinarily fast or anything, but it pulled people in, you know. But it does take you having to wave a flag of, hey, this is a serious movie or something, and some people want to be entertaining quickly. I'm really right in the middle, Matthew, because I admire people and I've done it myself to go on a talk show. Hey, we got to keep movie theaters alive. We got to keep this genre alive. And another part of me feels like if people want to see it, like Barbie, like Oppenheimer, they're going to go see it and go out of their way to be loud and proud about it. And I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive. Even though no one cares about this anymore. All respect to the ballet and opera people out there. I just lost 14 cents in viewership. But crazy shots. That's not a shot. I hear what you're saying. Yeah, yeah, Yeah, yeah. I don't know, it's interesting. I was thinking about, like, if like the creator of GTA 5, like stood on stage and was just like, we are 10 times the size of the baseball. Baseball. But also like the movie industry, like the, the video gaming industry has been basically 10 times the size of the. Of the movie industry for. You mean the movie theater business? No, like, like Hollywood, like gross production. Yeah, totally. I'm almost positive. Not. Not 10 times the size of your streaming platform. Yeah, maybe streaming. That includes TV shows. And then do you include mobile games or not? That's a big question. But the video game industry is definitely bigger. Raghav in the Twitch chat From deep Nvidia CEO just said he sees 1 trillion in revenue through 2020. Bring down the gong Bring down the mallet. Congratulations. It's amazing. Thank you. And we have our next guest in the Restream waiting room. First, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And we are joined by Kevin from Epic Gardening in the Restream waiting room. Let's bring in the CB Pro jump. Kevin, how are you doing? What's going on? What's up, brothers? How you doing? Good to see you, brother. Thanks so much for taking the time to join the show. First up, we got to talk about that tank. Yeah. What's in the tank? What's in the tank we've been talking about? I'm breeding rare Costa Rican tree frogs in this tank. No way. Yeah. They're endangered, okay. Oh, they're endangered. Okay. What else is. What else is special about a random? And you're planning to release them in all 50 states. States once you have enough. This is the goal. This is the goal we're always scaling over here. Yeah, of course. I love it. I love it. Is it challenging? Like, how much of your time is devoted to that particular tank? Almost none. Almost none. I just need to make sure that they're fed. Yeah, that's cool. What do they eat? Yeah, they eat crickets, which I'm breeding in that little tank right over there. You can see that. Yeah, we're breeding the different trophic levels over here, for sure. Okay, and then. And then. Do these frogs have use in your garden? Is it purely just for fun? No, I'm just branching out to flora or to fauna now, I guess. Yeah. Here at Epic, you know. Okay, well, first time in the show, so I want to kick off with your backstory. I want to know about the decision to start making content. I feel like that's always an interesting origin story. Like, when did you think, okay, I need to make content, dude? I mean, I'm an Internet OG, so I was on GeoCities. I was on Angel Fire back in the day. I was on Angel Fire, too. Yeah. Anime tutorials, you know, So I don't know what it is. I think genetically I'm designed to make content, but for Epic, it was really a calling card for. Remember when you used to design WordPress websites back in the day? Like, when people actually paid for that service? I used the.

He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets, he says. I literally think all the marks are wrong, Apollo's John Zito said of private equity and previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives of the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar, and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary, according to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public. I don't know. Calling people in private markets arrogant is crazy. I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold call by him, but we'll see what evidence he has back up that extraordinary claim. He blamed the media for creating a frenzy around private credit. Obviously we're in the middle of a private credit party, apparently, if you do credit well. Honestly, I would say we don't understand private credit well enough to really put everyone up into a frenzy.

Include mobile games or not. That's a big question. But the video game industry is definitely bigger. Raghav in the Twitch chat From deep Nvidia CEO just said he sees 1 trillion in revenue through 2020. That's a gong. Bring down the gong. Bring down the mallet. Congratulations. That's amazing. Thank you. And we have our next guest in the Restream waiting room. First, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And we are joined by Kevin from Epic Gardening in the Restream waiting room. Let's bring him in. CB panel. Jump. Kevin, how are you doing? What's going on? What's up, brothers? How you doing? Good to see you, brother. Thanks so much for taking the time to join the show. First up, we got to talk about that tank. Yeah. What's in the tank? What's in the tank we've been talking about? I'm breeding rare Costa Rican tree frogs in this tank. They're endangered. Okay. Oh, they're in danger. Okay. What else is. What else is special about a rare. And you're planning to release them in all 50 states once you have enough. This is the goal. This is the goal we're always scaling over here. Yeah, of course. I love it. Is it challenging? Like, how much of your time is devoted to that particular tank? Almost none. Almost none. I just need to make sure that they're fed. Yeah, that's cool. What do they eat? Yeah, they eat crickets, which I'm breeding in that little tank right over there. You can see that. Yeah, we're breeding the different trophic levels over here, for sure. Okay, and then. And then. Do these frogs have use in your garden? Is it purely just for fun? No, I'm just branching out to. To flora or to fauna now, I guess. Yeah. Here at Epic, you know. Okay, well, first time in the show, so I want to kick off with your backstory. I want to know about the decision to start making content. I feel like that's always an interesting origin story. Like, when did you think, okay, I need to make content, dude. I mean, I'm an Internet OG, so I was on GeoCities. I was on Angel Fire back in the day. Angel Fire, too? Yeah. Anime tutorials, you know, so I don't know what it is. I think genetically, I'm designed to make content, but for. For Epic, it was really a calling card for. Remember when you used to design WordPress websites back in the day, like when people actually paid for that service. Yeah. I used the blog as like a calling card or a digital business card for, like designing websites for. For local businesses and then just kind of kept plugging along with it and adding different platforms. And here we are today. Yeah. What about the first YouTube video? Like, what was the backstory behind choosing to go to YouTube? Choosing to go to video. It's a big lift for people if they're on substack or they're a writer and they don't know how they're going to do in front of camera. First YouTube video was 2013. So it was a long time ago. And ironically, back then, I mean, SEO and blogs were kind of the thing. Yeah. And so for me, the first YouTube video. Maybe it's the second YouTube video. You can see me using a screen recording app, reading a blog article. Just literally reading the article. Yeah. And with the hopes that people would watch that video and click the blog link and I would make money off of the advertising on the blog. So it's a completely backwards logic to today, of course. Yeah, yeah, yeah. But then obviously, discover YouTube is a far better platform, especially these days. So what was the. What was the flow of traffic over time? Were you able to reroute blog viewers to YouTube or did the algorithm eventually kick in because you're pre algo feed, right? Yeah, yeah, I think so. Right. I mean, I think it was back then, if you subscribe to a channel on YouTube, that subscription would just show up, which was a beautiful time. But no, I think every platform, as you expand every platform, you think, like, okay, well, I can get someone from this one to that one. It tends not to work. You tend to have to just play each platform for what it is. And so, like, YouTube became its own thing insta. All the other social media platforms have become their own thing. Now. How do you think about, like, serializing content, creating through lines, like the initial formats? Like, what was the actual development of the playbook that you ran on YouTube? On YouTube, I think in the early days, because remember, like, I'm 13 years old as a YouTuber, which is like two YouTuber lifespans, I think last about six years or so. And so back in the early days, it was just pure SEO, especially for a gardening channel. It's like, hey, how do you. How do I grow basil? How do I grow tomatoes? How do I prune tomatoes? These. These days, those videos have all been made either by me or someone else. Sure. And so we've had to come up with formats that, that work repeatedly over time. So for us it's great. I mean it's very seasonal business. So in March, what to plant in March, in April, what to plant in April or you know, in June, how to take care of your garden in June, that kind of thing. And then also coming up with like formats that are a little bit more high effort but tend to do better, like garden makeovers or garden tours where you actually have to go somewhere. Sure. But it's, it's easy to kind of like bulk, bulk those into a week and produce them. What did the journey look like of transitioning from media into actually making products yourself? Because that is an idea that is, at least in the venture world. People talk about just a very obvious transition. You just content to commerce and yet there's actually so few creators who have made that transition. Well, actually created products that go on to have equity value. I mean we, we, we've people bring up all the time, oh, you guys have this audience in, in tech, you should create, you know, software, various products for, for the audience. And our answer has always been, look, if we do that, we're competing against someone in our audience who is spending 100% of their time on that, on that business. And a sponsor. Yeah, and they could be a sponsor, but more so like I don't want to compete with someone in our audience that gets to spend 100% of their time on something when we can only spend like 10% of their time on it. Like they're going to smoke us. But I think in what you're doing, like very, very niched, very, very niche down. And maybe the companies that you're competing with are not like they can't go out and get $100 million of funding necessarily right away. But talk about that transition and how it's evolved. Yeah, yeah. I mean I think up until 2019, Epic was just media business and that's it. And it would be Google Ads, it'd be YouTube ads and maybe some brand stuff here and there. And I think in 2019 we did, out of just that pool, a quarter million in revenue. And then that was the year I decided to do product. And so the whole logic being I can't really control any of those three streams of income. Like traffic goes down for one reason or another, all of those go down commensurately. And so I thought, okay, well what can I sell? And the beauty of having content is that you kind of get like a pre validation engine for what you might want to put out there. And so there was this raised bed that I had. It's just like a metal garden bed that had been sent to me and I was like, this is the thing I get asked the most about, so I'll figure out how to sell it. I didn't even know who gave it to me initially. So I tracked down the manufacturers Australian company and I just kept emailing them every quarter. I was like, can I sell this? Can I sell this? They said, no, no, no, no, no. They eventually said, yes. I think I had 70 grand in the business bank account, 40 on a shipping container. I knew nothing about E Comm. So what I thought I would do is this is the most crazy, stupid E Comm logic of all time. But what I thought I would do is bring it into the port of San Diego, which does not take containers. So that, that was already a no go. It goes into the port of Long Beach. I thought I was going to go up and get it. Like me at the port, driving, driving the container down. I have a container here. I'm just picking up. Yeah, just like hauling it down. And then I was looking into Costco Self Storage to like rent that, unload the container and like get like some sort of satellite Internet to print the orders. And I talked to a couple of friends and they were like, yeah, have you heard of a third party logistics company, just ship it there and you know, just so stupid. But that's how little I knew at the time. And so what happened is made the order, got it on the water, made an Instagram story and said, hey, all these beds you guys keep asking about, they're here now. I have 550 of them. They sold out in two days. Use that cash to buy another container. Sold out. That out in two days. So like by the end of the year, I think we did quarter mil in just that. So the business doubled. And then of course, setting that up before the global pandemic was insane. So we went from 500 to like 2.8 million to 7.1 million the next year. And then raise. Raised a Series A. But yeah, I mean, immediately I was like, oh, this is obviously the actual revenue driver behind this business at least, Which I agree, like, a lot of media businesses don't have that easy plugin. Yep, totally. Yeah. What was the team like before and after this transition? Did you have to hire business people? How did you, like, how did you feel your role was changing? I mean, we've had Doug demuro on the show a few times and he was like very happy to hire a CEO to sort of run cars and bids and go back into content mode, do podcasts, which grew a ton. But every creator has sort of has a different journey as they evolve the business. Yeah, it's so weird because I run into Doug all the time at the coffee shop down the street, so. And we share the same investors, but yeah. So up until 2021 at tail end is when I raised the Series A, it was me or contractors. So it was me, editor, a writer and an assistant. And that was it. And we had. Did. We did about 7.5 million that year, mostly product sales at that point. So it was like, wait, wait, wait, wait, wait. So you have four contractors, but all those contractors are on the content side. But mostly it was. Yeah, so I was doing all the commerce stuff. So you have like. But you're single, you're a single product at this point. You've just made the best. It was single product. I'm just going to sell. I made. And you didn't have to develop the. I'm sure you made changes to the product. I didn't make the product. I mean, I think that's the biggest thing here is I did not make the product. You're living the drop shipping dream. Like literally it was thing that everybody gets sold and then it doesn't actually work because it was crazy level drop shipping, I guess you could say. Except for, I mean I, I owned the inventory, I brought it In, I had a 3 PL, like, so it wasn't true drop shipping. It's just that I didn't invent the product. It was a distributor relationship. Eventually, of course, we've started inventing products and you know, we scaled, I think from December 21st to December 22nd from four people to about 90 because we used some of the funding to buy a seed seed company that had 60 people. So yeah, that was a pretty crazy transition. That's. And talk about that the buy versus build decision on the, on the seed side, because I'm sure you had opportunities to do both. Right? So. So with seed, it's almost always going to be a buy because the infrastructure to actually like acquire seed. We sell almost 800 varieties of seed, vegetables, flowers, herbs. It's nearly impossible to scale that really quickly. If you have like buyers relationships, the buy orders are out a couple of years. You need like pretty specific infrastructure to actually like germinate and test those seeds, to pack them appropriately. I think there's like three or four companies maybe that sell the packing machines and they're all in like Germany. So some German guy will fly over and like fix a machine for you. So yeah, I mean and plus let alone like we bought the brand of this, of seed that I actually started gardening with back in the day. So there's like a heritage sort of story angle there that worked out really well. Yeah. What about the like your role shifting as you bring in those 60 new people? I imagine that they had a leadership team at a company of that scale. How are you interfacing with them? What does your role look like then? Yeah, I mean the first year or two was like all out madness. It was like whatever I could do at any point in time. So like still be the face of the content and architect that but you know, hiring, scaling, all sorts of ops, types of decisions. Now we have a president similar to Doug set up which is extremely, extremely helpful. He's ex chief growth officer at GameStop back in those crazy days. So he's got some pretty, pretty wild stories. Yeah. And with the seed brand the founders wanted to leave and so we had like this little holdover position for them and she kind of coached our leader in and they were just ready to go. And we can always call on them if we need them, but we don't, we don't really anymore. Yeah, that's great. Talk to me about seeds as a particularly good e commerce business. I imagine like when I, when I think about the worst e commerce brand, it would be like I sell a gallon of water. You know, it costs $20 to ship it and people buy it for a dollar. Low margin seeds. It feels like great e commerce product that maybe people just needed to be educated about. But was that your experience and what was it like actually scaling up? Yeah, I mean I think like those original products, the raised beds, like I didn't have to invent them. Right. Which is great. But every, by every other metric they're not a good econ product. Totally. The lightest one is 20 pounds, the heaviest one is 60 pounds. And then you're also, you're charged on dimensional weight of the shipping as well. And at the time my, my 3 PL was out of like Thousand Oaks. So I'm shipping from SoCal to the whole country a 60 pound box, which is just terrible. The beauty of that time is that I was charging shipping which is kind of unheard of these days and I had no customer acquisition costs. My customer acquisition cost was actually negative because I was getting paid to make my YouTube videos, you know what I mean? And that's what was selling it. And so I remember back in those Times pre. Pre funding, let's say kind of like laughing at all the DTC ecom Bros. Because I was like, you're running paid ads like clout. And now I'm like, okay, I understand the model a little better. But yeah, I mean once we got the seed brand, that's a primarily wholesale business. And so when we looked at it, I would say about 15, 20% of the revenue was direct to consumer and they had not focused on it. And so we've tripled D2C just by saying we own the business basically. We haven't done like a crazy amount of improvements as far as like DTC goes. We just, we just actually paid attention to it and plugged it into content. But you're right. Yeah. The gross margin on seeds is quite good relative to everything else in the gardening space. Yeah. How do you think about the transition from. I mean it sounds like you're actually doing the backwards transition. Most of like the D2C bros start online and then eventually they realize that. Okay, well I feel found the efficient frontier of cactl TV on meta and Google. Now it's time to go into retail. And then the whole company needs to pivot. They need to hire retail salespeople. Are you going in one direction or both directions? I've always wondered about the retail side of the business. Yeah, yeah. I mean I think the logic of the seed brand logic to me, I think there needs to be like a first order logic of buying something and that needs to be true. And then the second orders can be like very beneficial and may or may not play out. But for me this, the logic was like what we just talked about. The seed margins are very good and it's actually the only item in gardening you literally need every year. Every other thing you technically could get away with not buying again, like a raised bed or something like that. And so there's a repeatable addition to our business that we now have. But yeah, I mean this sort of like second order thoughts of, of buying the seed brand was can I introduce the raised beds, these seed trays that we developed to the wholesale network? Because that is very, very hard to build out. We're in 75% of all independent nurseries in the country, which it would be different if like we had Home Depot or Target or something and we could just say take this line instead we have reps that can go out to like 5,000 stores and say do you want this line? Which if we can get penetration on like some of those harder goods, then that's a Huge benefit that could play out for us. I have this thesis about creators that do that launch products is that they typically underrate the number of B2B buyers in their audiences. And you might not, you might think, okay, I'm selling a protein shake, I'll sell it to the consumer. But you might have like literally someone whose job is to buy the next protein shake, who works at Target or Walmart and they might be familiar with you. Have you had any of those experiences? Has that been advantageous or you know, is this a unique, unique industry? It's actually really weird because like the advantages you get, let's say in like pre validating a new product you might launch by teasing it in content and sort of seeing early demand. You actually get that to some degree with the wholesale relationships. Like we're in, we're in about 1300 petcos now. And I would say the sole reason is because the major buyer at Petco's just been an epic fan for a long time. So we were warmed up, you know, I don't have to go chase that down and prove it out. We're talking to Walmart for some stuff. Hopefully that comes to be. But it's a similar sort of way that relationship started too. So I think like the content angle, if you can convert it, you have some interesting doors like kind of automatically open. Do you think you have the most AI proof business in the world as we talk about like one, you're obviously not like trying to sell like you know, vertical software, but, but two, even, you know, the, the Citrini piece pointed out that there's a lot of like AI proof businesses where if demand gets destroyed because your buyer is no longer making 250k a year to do some email job, like your business might be fine, but maybe, maybe there's less demand. But I feel like even in, even in these like AI doomsday scenarios, like I probably, you know, if I, you know, lose my job, I still probably want some seeds, put them in the ground. Oh, I saw that anthropic piece that came out saying like which, which industries are the most vulnerable and I saw groundskeeping at a near zero. Near zero, which is, you know, gardening is just a, you know, a recreational version of groundskeeping. So I think we're fine. Yeah, yeah. How are you, how are you using AI? How are people using AI in gardens? Like I can imag taking a picture of something happening in your garden and just being like, how do I fix this? Like a lot of, a lot of ways that it could be Very useful. We have that. Yeah, we have that. So what we did is we launched this membership program that comes with commercial benefits. You get like 10% off the store, free shipping, free returns, which is great if you want to buy like a couple seeds here and there. And then we paired that with an educational sort of side because we have more or less the biggest gardening audience on any of the platforms. So we trained a model just on our own internal content and then like licensed databases of let's say plant facts or weather or something like that. And so if you ask it a question or send it a picture, it'll give you the answer that the closest answer you could get to what we would actually say, not just like what GPT or Claude might say. And then it'll kind of funnel you to live support if you want it, so you can get actual humans too. So it's kind of like a two tier thing and then we're just using it like along with anyone else that how you'd use it inside of a company for operations and stuff like that. What about on the content side? Are you finding it useful for scripting or thumbnail development or prototyping or sort of like layout? Anything that. I think it's good, John. It's Game changer. It's Game changer. Yeah. I mean, I think it's like the way we try to use it for content is like you're, you're really good first draft that you would normally have to spend however long to script out. The beauty of gardening, I think is it's so bespoke to like a particular individual's approach or a particular geography. That AI is not really crushing that right now, nor do I really want it to be. But it's really good for first drafting a lot of different things in content. Yeah, yeah. How do you recommend I fall in love with gardening? I grew up, my parents basically forced me to do a lot of weeding, a lot of mowing, a lot of just random stuff around our yard. I had bad allergies at the time, so I would come out of that and be like destroyed. And so I, I've not, I've had zero desire to, to get into gardening as an adult, but I feel like I just gotta find the right wedge products. Yeah. So is it, you know, raspberries, tomatoes. Look, I mean if you pick, you pick the crop that you are the most excited to eat and cook with and you grow that, you know, so if it's tomatoes, I mean, I'll send the technology brothers a and Some seeds, no problem. If it gets you in the garden, fantastic. I love it. Happy to support you. Just tell us what to get. Yeah. How are you thinking about the interaction between the creator, economy, YouTube content and Hollywood? We've seen Mr. Beast is all over Amazon prime now. I could imagine you doing content with more legacy institutions. What's your philosophy around those distribution channels? You know, so the two things we've done that are kind of tasting that world is we have a Samsung fast channel now that we've licensed 200 hours, hopefully more soon. And then we just launched last week an eight episode series on Home Depot's YouTube channel. So kind of like a co produced series, not a show, like not on streaming. But that's coming around too, I think. I don't know, I mean I think that if you're Jimmy and you can get a massive check to do something on prime, like why would you not? Right. But a lot of us on the smaller scale or like maybe industry wide big, but not like global big, the fat, the fast channel deals are looking really good right now. The, the sort of shows, if you can brand, even if it's just a YouTube series as a show versus just like a video or a series of videos, it seems to be pretty palatable to advertisers these days. That's good. Which is kind of interesting because like fundamentally it's just a list of videos. There's nothing really different about it. But if you call it a show, like Michelle Curry, challenge accepted. I don't know if you know her. Oh yeah, yeah, she's great. That, that is very much like a show on YouTube and it plays really well for, for those types of networks. Yeah, yeah. And eventually you, you have the seasonal element that you were saying. Like eventually you can be like, here's 100 hours of just April focused gardening content. And that's like, that's super powerful because the content is evergreen, that the plants and the earth and all these things aren't changing really in any sort of meaningful way year over year. That's such a funny mind shift because I don't know if you've had this experience, but it feels like playlists on YouTube like never really got what they deserve. Did you feel that way too? Right? Yeah, yeah. I mean playlists, like maybe back in the early days you would like crank through a let's play video game series or something like that and just let the playlist run exactly. These days I told the team, actually, I was like, look at every playlist we have, prune them down and then like bucket them into more conceptual shows rather than like this is my grow Tomatoes playlist. It's more like, you know, so, so that's, that's what we're trying to do right now. Yeah, like with Doug you'll see like, you know, car reviews or like Listicles and like it's more the structure. But yeah, that like you could imagine. There's also the question of like how set up is Hollywood to work with someone like you? Because if, even if they're like, yes, like we want you to do a full season on hgtv, but we're going to need to pull you away from everything else for I think the corridor crew guys went through this a little bit where like the, the numbers just never matched up. Like they would get bigger and then Hollywood would get more interested. But then the opportunity cost of taking six months off to do a real Hollywood movie just happened to me. Yeah, this happened to me in the pandemic. So 2020, it was June 2020. I did a deal with Chip and Joanna Gaines, then burgeoning network Magnolia. I think they were taking over DIY at the time. And it was supposed to be this transformation show. You go back if this beautiful sort of thing. And obviously Pandemic kind of hampered that. I had just bought a house. So I pitched this idea of I'll just build out this house and we'll show you and we'll go through that. So it was like 45 days straight of hardcore filming, like 10, 10 plus hours a day trying to get this done because there's a skeleton crew. And 2020 of course was the year, I think we started that year at 180k on YouTube. I ended that year at over a million plus another channel almost at 100k. And so if I take those 45 days and just calculate, let's say I was making call, even just 15 more videos, it would have been not only more money straight up, but more sort of brand value to the business to just make the YouTube videos. And I think that's what all the creators are running into. Yeah, yeah. Have you, have you ever gotten tempted to do any of the like selling the actual end product? There's been a number of venture backed companies that are like, you know, trying to make the perfect strawberry or any of these kind of vertical farming things. I always wanted somebody to do one of those like Berry company, one of those like, like butcher's box style thing. But give me a live video feed from the ranch so you can actually, you know, if you have this like real time, 247 idyllic ranch and then, and then you're, you're able to like low tam there. But I mean, look like it's hard enough shipping seeds around the world and shipping hard goods that don't expire. I can't imagine how hard it would be doing perishables. I don't. I would, I would never want to do it, honestly. Yeah, yeah. How are you thinking about product expansion? You know, if you go to the nursery, there's so much there. There's certain things that you're equipped for, you're operationally set up for. And there's other stuff that's maybe better content, you know, could, could, you know, be marketed, but might be an operational challenge. How do you assess, like new chrome hearts, Epic gardening wheelbarrow. I would do it. I would. Yeah, sure, why not? You know, that'd be fun. I mean, look like for us there's a lot of room to run and seed, there's, there's like tens of thousands of stores we're not in. Just on our botanical interest line. We launched an epic gardening line, which is like a guaranteed to work sort of beginner's line. Maybe that goes through to big box. Because two thirds of gardeners spend their first dollar at a big box. We're not in any of them. Yeah. And the mission of company is to help people grow anywhere they are. So if that's where they walk in, we want to be there in some way. So I think you can, you can run this business quite, quite a bit further just on seed alone. Sure. And then we're architecting the rest of the product strategy around that. So our second best selling line of products is seed starting trays and equipment and lighting. And then from there it's raised beds. I think we probably do something in soil or fertilizers next. But again, like, do we own a fertilizer and soil mixing facility? No. And like, do we just want a white label how. Right, yeah, yeah. On. On fertilizer. Is the fertilizer. Broader global fertilizer crisis because of the straight. Is that going to trickle down to everyday gardeners or they're not. It doesn't really matter for them if they're, if the price even were to go 2x, they still don't need enough product. It probably is fine for us. I don't know. I mean, I think it's way more a problem for like industrial agriculture. I think for us we're probably fine. Yeah, yeah. Very cool. Well, what about tools? Linus Tech tips has a screwdriver. I know. Yeah. I was just hanging with their CEO and I don't know, you might even see an LTT Epic collab at some point. That'd be great. Yeah. Yeah. He's the king of collabs. Well, thank you so much for taking the time. Great to finally meet you. Absolutely. Love congrats on all the progress, everything that you're doing. And hello to Doug Demurrer this year. I'll give. I'll give gardening another shot. Want to see it now that. Now that we have Kevin. Kevin. AI. Yeah, I'll hook you guys up. Don't worry about it. I got you. Fantastic. You're the man. Talk to you soon. Great to have you. Take care. Let me tell you about Restream 1 live stream, 30 plus destinations. If you want a multi stream, go to restream.com and I believe we have our next guest already in the Restream waiting room. Paul Cunningham is the dog healer. And now he's in the TV Pin ultra. Paul, how are you doing? How's it going, Jess? What's happening? It's going fantastic. Great. I don't know if it's early or late, but thank you. Or is it shockingly early? The sun is just rising.

Agent published this this morning. To me, the second I saw that, I started reading it. It felt like taking a double scoop of C4. Is that a pre workout? Yeah. You never. I know the can. I didn't know it was a. You never dabbled. What was the one that we. I'm more of the gorilla mind one. That's the one that I. Many people have said you have the mind of a gorilla. Yes, yes, yes. For more plates, more dates. You're a gorilla in sheep's clothing. I think that's literally workout that I have, although I don't use it that often anyway. So you got pumped up. I got pumped up. Ben writes, there's a weird paradox in terms of AI prognosticization. Prognostication. Prognostication. That was a good, good effort, Jordy. On one hand, there's so many requirements. There's just so many words. What are the requirements for having a podcast? Like knowing how to say words. No taste. I mean, yeah, ultimately there's a lot of words that you. When you read them. Yeah. You're just like, oh, yeah, you can just do it. And then you try to rip it. On one hand, you don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out? To be foolishly optimistic. At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I very. I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026, however, on the morning of Nvidia GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gun going. He writes LM paradigms over the last couple of weeks, first in the context of Nvidia earnings, and then last week in the context of Oracles. I've talked to you about 3 LM inflection. I've talked about 3 LM inflection points. Yeah, I'm not going to go through all of these. He goes, chat. We've talked about this a few times. Reasoning models and then agents. And each one of those increases the demand exponentially for compute. So chatgpt01 and then opus as well as Claude Code and Codex. Codex. Basically getting to the point where tasks are being accomplished over hours and getting to great outcomes. And this is the interesting point, the decreased need for agency. The reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question.

Marc Andreessen chimes in, I can't load the post right now. We gotta go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32 inch MacBook called the MacBook Pro Ultra Wide, and it looked like this. I bought one and unlocked extreme productivity. And then it wouldn't fit into my backpack, so I had to leave it behind. Oh, no. This is sort of like on that other laptop that we saw. They should honestly make this. Should this should. I mean, walking around looking like maybe you could put skateboard trucks on it. Yeah. That you could use it as transport. Yeah, it's more of like a snowboard build that you like carry over your shoulder like this. Or surfboard. You know, people throw it on the top of your car like that, three fingers. Why you don't put a surfboard on the top of your car? Yeah, I mean, real ones don't. Oh, what do they do? They put it inside the car? Truck Batter inside truck. Yeah. I don't pretend in the LA area, it's. You can, you can clock if somebody's actually a surfer or not just by the way they go each with their board. Okay. No, but I think. Yeah. Throwing it under your arm, having some trucks on it, skating, being able to get where you need to go. I like the Ultra Wide. What if. What if they're driving a Huracan Sterato? Where would you recommend that they put their surfboard then? Jordy Stirato. I could make exceptions.

On the AI versus dog cancer. What happened so late Friday, there was a story about an Australian tech entrepreneur named Paul Coyningham reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Coynningham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's true. It's great to hear that. Australia. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Queen Ham used ChatGPT to brainstorm ways he could help. And he did an interview on this and here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark. Embark. DNA. You could do it. Okay, well, anyway, he went to a university, probably for a good reason, probably got good data. He said, the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 kilometers down the road. You can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Coynham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of, like, how much was this Cure my dog cancer. One shot. It don't make mistakes. I don't think anyone's saying that. But very quickly, there was like an incentive to. To amplify this into the hype, this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. So once the DNA sequence was produced, he ran it through a whole bunch of custom different data pipelines to find those mutations, and then used other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA institute, which used Cunningham's data, crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us and we made a little nanoparticle and it's democratizing the whole process. They said, this is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Corningham and his team administered the vaccine to Rosie, which was affected. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have a little bit. The question is, is it runaway, is it bad, is it terrible? And then it's hard to just snap your fingers and cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X, the news of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog. Air horn for Rosie. The air horn for the dog. That's great.

Happen. So late Friday, there was a story about an Australian tech entrepreneur named Paul Coyningham reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Coynning ham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's true. It's great to hear that. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Cunningham used ChatGPT to brainstorm ways he could help. And he did an interview on this. And here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23 in me for dogs yet or something like that? Who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark, embark. DNA. You could do it. Okay. Well, he went to a university, probably for a good reason, probably got good data. He said, the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 kilometers down the road. You can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Cunningham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of, like, how much was this cure my dog cancer one shot. It don't make mistakes. I don't think anyone's saying that. But very quickly there was like an incentive to. To amplify this into the hype, this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. So once the DNA sequence was produced, he ran it through a whole bunch of custom different data pipelines to find those mutations and then use other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA institute, which used Cunningham's data crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us and we made a little nanoparticle and it's democratizing the whole process. They said, this is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Corningham and his team administered the vaccine to Rosie, which was affected. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have a little bit. The question is, is it runaway? Is it bad, is it terrible? And then it's hard to just like snap your fingers and cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X, the news of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog air horn for Rosie. The air horn for the.

Let's just roll, right? Market clearing order inbound. Come get up. You're surrounded by gentlemen. Thank you. Strike 1. Strike 2. Activate. Go. Golden retriever mode. Online. I see multiple journalists on the horizon. Stand by, Founder. You're watching TVPN. Today is Monday, March 16, 2026. We are live from the TVPN Ultradome. The temple of technology, the fortress finance, the capital of capital. Let me tell you about ramp.com time is money save. Both easy use, corporate cards, bill pay, accounting, and a whole lot more all in one place. What a massive week last week. Alex Karp going back to back with Travis Kalanick. The reactions to the Travis Kalanick interview was phenomenal. I was reading them all weekend. I was still emotional the next day. Yeah. And there's something I posted this on one of those clips that someone just shared was like, this is a great clip. And I was there because you're in the moment and you don't realize barely. Do I reflect too much on different interviews. Cause there's always the next day of interviews. But watching some of the clips back, Guillermo from Vercel put together that hour long. So good motivational video. Yeah, it was so good. I think that the Travis Kalanick mindset has been missing. When he kind of left. Yeah, yeah, there was. There's been a Travis sized hole. Yeah. In the industry, in the culture. Yeah. And to see him come back and in, you know, 45 minutes, basically just give the advice that I think like everyone that's building in some way can benefit from. Not everyone is going to be Travis, but there isn't anybody out there that's done what Travis has done that is kind of like preaching that. And I don't like listening to like founder porn content personally. It's not. It's not appealing. But when it comes from Travis, it is just another level. Yeah. Like the right message at the right time. Yeah. Especially the thing. The thing that I was kind of pulling on is like right now, like there's a lot of easy money everywhere. Right. There's teams that have built nothing that can raise between 50 to a billion dollars at times. And his feedback on that, his point of view was like, okay, is capital really a constraint in your business? How much does it matter? How much is it going to matter in terms of the competitive dynamics of your market? And if it matters. And in a lot of these AI categories, it does. If it matters. And it was easy, that means you didn't go hard enough. Yeah, that was the best line. And that was like the best line like if money matters, as we all agree. So you raised a billion. Wait, why didn't you raise 2 billion? If money matters, why didn't you raise 3 billion? Yep. Oh, it was easy. That means you didn't go hard enough. Yeah, I mean, that's somewhat the subject of what Dylan Patel was talking to Dorkesh about on the Dwarkesh Patel podcast. Fantastic show, by the way. Fantastic episode about this, like, you know, being risk, on being aggressive. And Ben Thompson wrote about that today, you know, through a different lens, talking about, you know, are we in a bubble? Maybe, but like all the numbers are penciling out, so go, go, go. Like now is the time to scale. And yeah, it's. It was fascinating hearing it from a completely different perspective at the perfect time. But I really. Yeah, that was a great, great interview. That was personal highlight for sure. Building TVPN for sure Friday. Yeah, that was great. It was, it really was like, like the conversation that we set out to have because, I mean, he mentioned he's leaving California, but we're not going to like get bogged down in like his political views or whatever like that. It's, it's so much more about the actual craft of scaling a business. And like, I think, I think we just nailed that and so that was really fun. Yeah. And the good thing is we have plans to do a show like that every single day of the year. Three days. No, unfortunately, it's not possible. Right. It's not very often that someone like Travis. Yeah. World historic founder, comes out of media retirement after almost a decade. Almost a decade. Yeah. So very cool. But thank you for everyone who tuned in. Thanks everyone who enjoyed any of the clips, saw whatever you saw of it. It was a really fun time. And if you care, if you want to work in physical AI and you don't see yourself in the Elon verse, I think that is one of your best possible bets. Sort of like an indexed approach to physical AI. Insanely hard for Elon, work extremely hard for Travis. Yeah, you're going to have to work hard one way or another. Exactly. But that's the nature of. If you don't want to work hard, there's probably a company out there that's competing with Travis or Elon in physical AI. You could work there. I just wouldn't put much value on your RSUs. It's rough. Anyway, let's pull up the linear lineup. Linear, of course, is the system for modern software development. 70% of enterprises Workspaces on linear are using agents. We have Kevin Espertu from Epic Gardening coming on to tell his story about scaling his YouTube channel. I think we have a lot to talk about. We always love creator economy stories. Paul Coyningham, the dog healer is coming to break down how he used AI to augment delay his dog's cancer. We're going to be digging into. We'll first go through what actually happened. I have some opinions about this. And then we'll talk to him to get his side. Then we are pulling our delayed lightning round. We went far longer than we expected with Travis on Friday. So we are catching up on our lightning round with Tony from Sunday Robotics. Then we have drew from 8 VC is a founding partner there. He backed Quince at seed. It's now a $10 billion company. The Quince founder is a little under the radar. Totally. But I wanted to get this story from the 8VC team, hear how they're thinking about it, and then a bunch of other folks joining. Yeah, looking forward to that. Well, let's read through Brandon Gorell's deep dive on the AI versus dog cancer. What happened so late Friday? There was a story about an Australian tech entre, Paul Coynningham, reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Cunningham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's true. It's great to hear that. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Cunningham used ChatGPT to brainstorm ways he could help. And he did an interview on this and here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing, who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark, embark DNA. You could do it. Okay, well, he went to a university probably for a good reason, probably got good data. He said, the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor. And you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 kilometers down the road. You can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Cunningham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of, like, how much was this cure my dog cancer, one shot it don't make mistakes. I don't think anyone's saying that. But very quickly there was like an incentive to amplify this into like the hype, like this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. Once the DNA sequence was produced, he ran it through a whole bunch of custom, different data pipelines to find those mutations and then used other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA institute, which used Cunningham's data, crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us. And we made a little nanoparticle and it's democratizing the whole process, they said. This is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Cunningham and his team administered the vaccine to Rosie, which was affected. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have a little bit. The question is, is it runaway, is it bad, is it terrible? And then it's hard to just like snap your fingers and cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Coyningham says the quality of life of the dog Rosie is much better now. So on X news of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog. Air horn for Rosie. The air horn for the dog. That's great. Turned into a heated debate on health regulation after biomedical engineer Patrick Heiser posted that quote, it is trivially easy, trivially, trivially easy to make a single MRNA vaccine. It's not hard. And Hank Green, a prominent YouTuber, issued something of a rebuttal, which we can go through later. A separate thread in the discourse is focused on the promise of LLMs democratizing access to medical science, with OpenAI President Greg Brockman quote tweeting the story with the caption, a small window into the opportunity of AGI. Well, Coiningham didn't literally cure Rosie's cancer with ChatGPT. As Stripe CEO Patrick Collison pointed out, it acted as a high powered search tool that ultimately helped his team get to an amazing outcome. Sort of. George Hobbs. We got to move the goalposts. I'm ready to move them. Moving the goalposts. I mean, where are we moving them to? It has to, actually. You have to be able to type cure my cancer. And then from your phone it just deposits a pill that you just take. Yeah, exactly. Is that what it is? It has to. Locally. Yeah. End to end. No, ideally. Ideally it would be not even a pill that you take. It can just create a video that you. The right pattern of light. The right pattern of light coming from. And sound. So the phone has light and sound and so the light flashes in your eyes at a certain rate. It rewires your brain and your brain decides to go kill the cancer. Yeah. And we've talked about this a bunch. I think, I think it would be helpful for the industry to refocus some messaging on not AI is going to cure cancer, but human humanity is going to use AI to cure cancer and do a number of other things. Right. And so the, the bar is not just like one shotting it with a prompt and it sends it to a lab and you get a, you know, some, some type of treatment in the mail. Maybe. We, we. I can imagine that in the future. Right. Something, something to that effect, but it is an enabler, it's a tool. Yeah. And this has allowed someone to become, not an expert in something, but to help somebody understand a process enough to go out and find the right experts to help them solve their problem. And I Think it's incredibly inspiring. So excited to have him on the show later. So there was a chemist who works in AI and biotech by the name of Ash Jogalikar, and he had a really good summary along those lines with a riff on Freeman Dyson's 2007 New York Review of Books essay R Biotech Future, which we should read at some point, in which in this article, Freeman Dyson argues that biotechnology will become small and domesticated rather than big and centralized. The full post is worth reading in full, and we might go through it, but the conclusion is particularly good. If AI continues to reduce the cognitive overhead required to navigate biological knowledge and assemble complex pipelines, the boundary between professional research and motivated individuals may begin to blur. That shift will require careful thinking about safety, governance, and responsibility. But it also carries an exciting possibility. Dyson imagined a world in which biological design might eventually become something like a creative craft practiced not only by institutions, but also by curious individuals experimenting at smaller scales. Yeah, I think there's the reality of cancer treatment, from my understanding is, and this was based on a late family member that had cancer and ultimately passed away. During the process, during his treatment process, which was around a year and a half, he was getting looks at different treatments that were promising, some of which he was able to do. Some he didn't qualify for just based on his personal situation, even though there was. There was a decent chance that it could have had a positive effect. Yeah. And that sort of the insane frustration that an individual feels or a family feels when they're like, hey, this, you know, if something's terminal or it's looking really bad, it's progressing the wrong direction. And there's a, there's a treatment out there that isn't. That is somewhat trivial to actually make. You just don't qualify for it. That level of frustration will eventually drive more individuals, I think, to do this. Right. And so there's definitely, definitely some, like safety. There's huge safety concerns. There's ethical concerns. There's. These are things that we have to work through. But ultimately, I just think there's going to be so much. There's going to be enough, like human energy and just overall desire to live that people will take risks that they wouldn't take for a bunch of other more sort of like trivial sort of issues. There have been initiatives with the fda, something around. Right. To try in certain scenarios, patients rights, sort of removing some of the regulation and allowing people to make decisions like that. It does feel like the FDA's stance might need to Change in this case, like they clearly have a role to play play currently and in the future where biotech becomes more democratized. But yeah, hopefully there's some good symbiotic relationship there with the broader biotech community as we get bigger. I have a similar story with someone who developed a rare illness and was able to go and read academic research at a very deep level. Didn't have a background in biotech or anything like that, but was able, this was pre AI, was able to read like every published research paper that was at all related to this particular illness and found the world expert in this particular disease, contacted the professor and the professor said, yes, you have the thing that I've been studying and I've only found five people or 10 people in my entire career that have this thing come down. I, I will operate on you. The operation happened, it was successful and it was fundamentally like a high agency person doing a lot of research. And if AI just acts as a search tool that democratizes that, you're going to get better results. So even if you're, even if we're not in like one shot in curing cancer, that just feels like making search easier, making research easier. Huge benefit. Yeah. Cunningham, the guy, Australia could have done a lot of this 10 years ago. He just would have needed to spend, I'm sure a bunch of time and yeah, that's the library. Everything you do, all these things you can do manually. You can, you can just get a guy for that. Yeah, you can get a guy or you, I mean you don't even need a spreadsheet. You can do, you can do this, you can calculate the math by hand. But these things speed things up. So it's, it's been a good time. Let's read through Ash's post. But first, first let me tell you about public.com investing for those that take it seriously. Stocks, options, bonds, crypto treasuries and more with great customer service. And let me also tell you about fin, the number one AI agent for customer service. If you want AI to handle your customer support, go to FIN A. Thank you for clapping. Tyler. How was your weekend, Tyler? It was good. Yeah, it was good. Yeah. Did you go to any data centers or are you. No data centers this weekend? I was in sf. Didn't you go to a pig roast? Yeah, that was on Friday. I was in El Segundo. How was sf? Is something big happening there? Does it feel like being in Wuhan in February of 2020? Something, something big was happening. Yeah, there was. I went to a Debate. Oh, you went to the debate. Okay, cool. How was that? It was good. Yeah, yeah. It was about the billionaire. Yeah, yeah, yeah, yeah. And did you go to the hackathon at all? No, I missed that. Oh yeah, I saw that semi analysis had a hackathon. The winners were crowned. Seemed like a lot of fun. It really does seem like the best time to go to hackathon just because what you can actually accomplish in two days is remarkable. Yeah, yeah. No, right, yeah. It's like people used to do hackathons and it'd be like after two days they'd be like, we have a landing page. And now it's like. And a cool idea. We created a hackathon simulator with mini games for everything and it's also making money. We need to give an update on TVPN simulator at some point, but it is coming along. The development has continued at breakneck pace. Yeah, we got to work on the rollout of this. Yes, it might be GTA 6 level by the time GTA 6 comes out. I think we can get there. We need a new graphics package. What do you think the actual path to AAA graphics is? Do you think we should rewrite it in Unreal Engine with ray tracing and insist that people only run it on gaming PCs or should we do some sort of style transfer on top of it? Yeah, I think the Unreal Engine is probably easier because you're just moving the code over. That shouldn't be that difficult. You probably do that in like a day or two. I think these things used to take like years. Like it took. It took Elder Scrolls like a decade to get to like Nintendo Switch. You're like, yeah, it takes a day or two. The real time render thing is interesting, but I think that's. It's just like expensive. That's the problem. We had. We had someone on the show that was doing it on Zoom over in real time. That was Descartes. Descartes, yeah. That was a cool demo. Yeah. So you imagine like that tech prompted with like make. Take this from. From like boxy. I would say we're at. We're above N64 level graphics, but we're probably more like Xbox 360 graphics and take us into, you know, modern day PS4. Yeah. I mean this is why I'm very excited about doing. Everyone's so up in arms about like, oh, the new like PSX isn't going to come out because you ain't the memory and people are like, oh, I don't want to play games in the cloud. Right. But if you're in the cloud, that means you can actually like access a ton of compute. Because like when you're not playing, when you're not using the GPU to run like the nice graphics, someone else can be. Yeah, you can actually get higher to. You can get access to much better like hardware playing video games and then also. Yeah, more and more iteration on the graphics. Like it should just be like live service model basically. Yeah. And if you get the Genie 3 model where it's actually generating on spot, like. Yeah, that's something you can really only do in the cloud. I'm excited. Jensen is doing his keynote at gdc. Should we pull up the live stream? We can. Yeah, let's check in with Jensen. Let me tell you about Okta first. Okta helps you assign every AI agent a trusted identity. So you get the power of AI without the risk. Secure every agent, secure any agent. And let me also tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Continue. Let's play it. What do we got? We got Jensen, institutional investor. These three people are deep in technology, deep in what's going on. And of course they have just a really broad reach of technology ecosystem. And then of course all of the VIPs that I hand selected to join us today. All star team. I want to thank all of you for that all star team. The leather jacket really has just aged so well. I also want to thank all the companies that are here. Nvidia, as you know, is a platform company. Mic drop. We have technology, we have our platform. Oh, by the way, everyone uses. He's mogging our merch. He is. And today there are probably 100% of the hundred trillion dollars of industry here. 450 companies sponsored this event. I want to thank you a thousand, I love it. Technical sessions. 2,000 speakers. This is 2,000 speakers. Wow. Every single layer in one. They're going to do more interviews than we've done all year. Structure in one chips for two days to the platforms, the models and of course the most important and ultimately what's going to take get this industry taken off is all of the applications. This really is the super bowl for semiconductors. It all began here. This is the 20th anniversary of CUDA. We've been working on CUDA for 20 years. For 20 years we've been dedicated to this architecture, this revolutionary invention. Simt, single instruction, multi threading. All right, very, very cool. Let's get back to the timeline. Let's go to Gemini 3 Pro. Gemini 3.1 Pro is here with a more capable baseline. It's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life. And let me also tell you about Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases and more. While Railway automatically takes care of scaling, monitoring and security. Back on the timeline we gotta go through. Wait, yeah, you can just save your dog. The beautiful picture here. You can just save your dog. It's remarkable. This is a heartwarming story. And it also, yeah, I really like how it reveals current AI capabilities, where things are the benefits and sort of the diffusion narrative. This is fundamentally a diffusion story, not a super intelligence story in my opinion. But let's go through Ash's post here. My take on the whole AI cures dog cancer in Australia is a very interesting story, but perhaps not for the reasons that are being noted. In 2007, Freeman Dyson published an essay in the New York Review of Books called Our Biotech Future. It contains one of the most memorable predictions about the future of biology that I've ever read. I predict that the domestication of biotechnology will dominate our lives during the next 50 years, at least as much as the domestication of computers has dominated our lives during the previous 50 years. Dyson believed biology would eventually follow the trajectory of computing. At first, powerful tools live inside large institutions, universities, government labs, major companies. Over time, these tools get cheaper, easier to use, and more widely distributed. Eventually, individuals start doing things that once required entire organizations. You will be the manager of infinite minds. You will have a million agents and you will also have access to the equivalent of a university lab filled with biotechnology equipment. Biotech will become small and domesticated, rather than big and centralized. This is very interesting in the age of AI because there's been this narrative of like, AI is a centralizing technology, it is very power law driven. But this is sort of counter to that. I don't exactly know how to piece those two things together, but it is interesting that his prediction was actual decentralization in this particular category. He even imagined genome design becoming almost artistic. Designing genomes will be a personal thing, a new art form as creative as painting or sculpture. Dyson's words rang in my mind as I read the AI cures dog cancer story. Much of the coverage framed in. I gotta say, it's very easy to imagine you in 20 years. I'm like, John, like, you gotta tell us your anabolic steroid and you're like, it's kind of a personal thing. It's kind of a personal thing. It's kind of like an artisanal process that I go through a sculpture, I'm sort of sculpting myself. I can't really, I'm sorry, I can't really share my stack with you, but it's a personal thing, so go and kind of figure out your own stack. You know, speaking of sculptures, I was walking around my neighborhood and I looked through this like, you know, gap in the trees into this like large lawn and I saw on this person's like front, front lawn, behind like, you know, gates and whatnot, just a full size statue of a man playing golf who I didn't recognize. It was like, it was not tiger wolves. I think it was the owner. I think it was the owner. I think the owner was like, I'm into golf. Or you know, like one of his boys got it for him, which is a hilarious gift. Getting someone a life size statue of themselves and just having it delivered. And then it's like, well, it's impolite for you to turn it down. You know, what are you going to do Anyway, the scientific pipeline involved here is actually well known. It closely mirrors the workflow used in personalized neoantigen vaccine research that has been under active development for years. The steps are fairly standard. Sequence the tumor, identify somatic mutations, predict which mutated peptides might be recognized by the immune system, encode those sequences into an MRNA construct and deliver them to stimulate an immune response. The biological targets themselves were almost certainly not new discoveries. I have been able to. I haven't not. I have been unable to find out what they are. But mutations in targets like kit, which are common, might be involved partly. Therein lies the rub. Since the hardest part of drug discovery, whether in humans or dogs, is target validation, the lack of which leads to a lack of efficacy, the number one reason for drug failure in neoantigen vaccines. The proteins involved are usually ordinary cellular proteins that happen to contain. Contain tumor specific mutations. Alphafold, which was used to map the mutations onto specific protein structures, is now a standard part of drug discovery pipelines. That's fascinating. The challenge is identifying which mutated peptides might plausibly trigger immunity. What is interesting though, is how the pipeline was assembled. Normally, this type of workflow spans multiple domains. Genomics, bioinformatics, immunology and translational medicine. And in institutional settings, those pieces are distributed across specialized teams, document sources, and legal and technical barriers. Navigating the literature, selecting computational tools, interpreting sequencing results and designing a candidate MRNA construct is typically a collaborative process. In this case, AI appears to have helped compress that process, pulling together data and tools from different sources. Instead of requiring multiple experts, a motivated individual was able to assemble the workflow, with AI acting as a kind of guide through the technical landscape. That is fascinating. Anyway, it's a longer. It's a longer post, but you should go read the thing in full. Patrick Collison also chimed in. He said, according to the story, the dog's cancer has not been cured. I think it's just 50% smaller, which of course is a win. But not using the term cure is always tricky, but it does go viral. Absent all regulatory and manufacturing constraints. We could not just synthesize magic RNA, RRNA cancer cures. The technology is very promising, but it's not any kind of panacea. Yet the emergent system, emergent system of system of regulators and manufacturers is indeed far too conservative and small. Small scale experimentation is much harder than it should be. More people should read the first part of the Raw the Rise and Fall of Modern Medicine. So it's interesting, Lee says chatgpt Cure cancer, Make no mistakes Biomedical engineering industry yeah, don't do this. It's easy and effective, but we can't make enough money off of it. That's ridiculous. It's surprising. G. Fodor says it's surprising how people are so blatantly talking past each other on this. The point is that the system of clinical trials is predicated on an assumption that a given drug will work on a cohort. What if there are lots of drugs that will only work on one person? So definitely a big desire and push for rethinking the system of clinical trials. If you're going to have personalized medicine, what does that mean? There's already a lot of people, biographers that do all sorts of stuff like this. Can't believe he wasted two cups of water to do this. Banai it is ridiculous. It's a great counterpoint to the doomers. What else is going on? Marc Andreessen chimes in I can't load the post right now. We gotta go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32 inch MacBook called the MacBook Pro Ultra Wide and it looked like this. I bought one and unlocked extreme productivity and then it wouldn't fit into my backpack so I had to leave it behind. Oh no. This is sort of like based on that other laptop that we they should honestly make this. They should. This should. I mean, walking around looking like maybe you could put skateboard trucks on it. Yeah. That you could use it as transport. Yeah, it's more of like a snowboard build that you like carry over your shoulder like this or surfboard. You know, people throw it on the top of your car like that. Three fingers. Why you don't put a surfboard on the top of your car? Yeah, I mean, real ones don't. Oh, what do they do? They put it inside the car Truck bed or inside truck bed. Okay. Yeah. I don't pretend in the LA area, you can clock if somebody's actually a surfer or not. Just by the way they go each with their board. Okay. No, but I think. Yeah. Throwing it under your arm, having some trucks on it, skating, being able to get where you need to go. I like the ultra wide. What if they're driving a Huracan Storrado? Where would you recommend that they put their surfboard then? Jordy Stirato. I can make exceptions. Okay. I like this. Dylan Patel said on Dwarkesh, The TAM for GPT 5.4 is north of $100 billion. But there's adoption lag. That's considered AGI as far as the Microsoft OpenAI contract is concerned. That's very interesting. Sam Carter says the reported 1 billion of profit is no longer the sole trigger for confidential IP research access. It reportedly includes an independent expert review. You were saying Joe Rogan would be on that, Andrew Huberman, the experts would be on there. You got to trust them at all times. Neo Vaughn. Maybe the funniest thing about that joke is that I actually would like to know that panel of experts where they deem AGI, because I feel like between all of them, they could chat with the chatbots and be like, it's not that good yet. Be very realistic about it. Yeah, they're not necessarily just going to be like, oh, I'm pumping it for whatever reason. I have this weird bias, whatever. And so it'll be very interesting to see how that, how the AGI definition plays out, because it does feel like we're close. I mean, Dario on Dorkesh was saying, like, we're near the end of the exponential, which is like, sort of crazy. It feels like, you know, Sequoia declared AGI. They're an investor in OpenAI. And so there's a lot of stuff. What do you think about the AGI? Do you want to be on the expert panel? I think I would say that we already reached AGI. It was maybe earlier, you called it like 30 seconds before Tyler Cowan did. I think I remember it was like 30 seconds before you came. You tapped me on the shoulder and you said it's here. And then we went and refreshed X and Tyler Cowan had come out. Yeah. I think realistically I'd probably say it was something like when the agentic harnesses came out. So stuff like Claude code. Sure. Where you can actually just tell it to build a project and then there'll be errors. It'll see those errors, it'll fix them. It'll keep working on it. Not reasoning models. I mean it's so hard. It's like on like math or something like this. Right. But those basically unlocked. Yeah. Now they can just do anything. Yeah, yeah. I mean the agentic thing was talked about for a full year and then it finally happened like in December and it was pretty broken up until then and then. Yeah, but I think like you can still just make like a very good case that like. Yeah, chatgpt like that was AGI like you can just ask question. It'll answer it. Yeah. If you'd never talked to an AI model before and you talked to that, you're like this. Okay, this is a person. Microsoft Excel 1985 AGI Jose Macedo says ultimate narrative violation from the Dylan Patel to our Keshe pot three years later, H1 hundreds are actually trading above launch price in secondary markets I. E. Negative depreciation. Yeah, that's called appreciation. Appreciation. I appreciate. I just want to go out and say I appreciate age 100 negative depreciation. That's. This completely flips the Michael Burry two year E waste bear thesis on its head. Yeah, I mean somebody's got to check on. Somebody's got to check on Michael Burry. It's such a different, it's such a different dynamic because I mean like the whole. There was a reasonable underpinning for GPU depreciation which was just look at 20 years of computer equipment history. It's like it all depreciates over like maybe five years, maybe 10 years. Some stuff sticks around but like they burn. Yeah. It's just interesting. Jose says core we probably benefits most from this. They have 250,000 GPUs and a $66 billion backlog. Depending where you think market was pricing depreciation margins improved by something around 40% which means 1 billion a year in additional earnings. Who knows where this stuff actually re rates or how sustainable. But great, great time to be a neolab. One of. One of the founding team members At Lambda was posting last week. Basically congratulations to everybody that booked out like GPUs on an annual basis in 2025. You're looking like absolutely brilliant right now. Obviously, Sam is starting to look extremely vindicated on all the deals that he did last year. So, Tomas, quickly let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB? Don't just build AI, own the data platform that powers it. And let me also tell you about TurboPuffer, serverless vector and full text search. Built from first principles and object storage. Fast 10x cheaper and extremely scalable. Tamaz says we've been growing a lot and are out of GPUs. This is Sam Altman in March of 2025. Katz over at Oracle says we are still waiving off customers or scheduling them out in the future. This is a situation that we have not seen in our history. Satya says you may actually have a bunch of chips sitting in inventory that I can't plug in. I don't have warm shells to plug into. Sundar says, what keeps us up at night? The top question is definitely around capacity. All constraints, be it power, land, supply constraints, how do you ramp up to meet this extraordinary demand? And sorry, quickly, between power, land, supply chain constraints, chips. You were saying that the, the, the takeaway. Your takeaway or your read on Dylan Patel on Dwarkash was that chips were the main. Yeah. I mean, not even like a read. Like he explicitly said. He's like, between power and chips. Chips. Chips is what's going to be the big bottleneck. Because at some point, like. Yeah, there's all these ways that you can actually like, maybe get like 10% of the, you know, US energy production to just like go to. Yeah. You know, where like at some point, like, okay, we don't have enough, like UV tools. Yeah. And like they're not building them right now, which means that they're not going to have them for at least three, four years. Yeah, yeah, yeah. This was Ben Thompson's. Like, TSMC needs to step up and spend more on CapEx. Their. Their CapEx guide is like a Capex guide for ants. Like a mere 45 billion or something. And it should be probably much, much higher. Yeah. But I mean, it even goes down to the, you know, the toolmakers below them, like really, like really deep in the supply chain. Yeah. At least what I got from Dilutel on that interview is that like, they still are not really that AGI pilled. They're not Expecting this kind of massive, you know, increase in demand to stick around. Yeah. Trey says a sign of taste is dabbling in the vintage GPU market. A1 hundreds. Yeah. Ampere v. You gotta go Volta. Back to Volta. Yeah, I mean I remember I was digging into that like chips versus energy. What's the big bottleneck? And I think we're using something like 50% of leading edge capacity like of the fabs that can make AI powered GPUs like GPUs that can run transformer based large language models. We're using like 50% of that capacity already. And then some of the leading edge nodes go towards like you know, Apple silicon chips that are maybe designed system on chip, something for a phone. And only like 1% of energy right now in America goes towards AI or less. It's like 0.1 or something. So you can reallocate and everyone just turn off your air conditioning. One more close the door. If the air conditioning lip Bhutan says there no, there's no relief as far as I know. No relief until 2028. Somewhat ominous. Keep reading what Tomoz says. What happens when your AI doesn't answer? Everything is in short supply. It's no longer just GPUs. It's power, data centers, memory, CPUs. If there's no relief for six more quarters, perhaps it's time to plan for a world where inference isn't freely available on demand. Inference prices which have been static will rise. Subsidies will be harder to justify. Enterprises will need to rationalize workloads deciding which teams receive state of the art models and which don't. Not every CRM update requires a trillion parameter frontier model inference rationing normalizes. Marketing receives this much sales, receives that much. Software engineers probably receive a lot more Constraint will be the mother of invention. Companies will optimize what they have, adopt open source where they can and likely move to smaller models for many workloads. This is a really cool take. I like this. It's also interesting to me is that not every CRM update requires a trillion parameter frontier model. That's another bull case for Hopper price stability going forward and less depreciation. Because if you can leave your CRM update workload where you're just going through and spell checking names and cross referencing data sources and pulling from email and dumping some notes, summarizing. And you can do that on a GPT4 class model instead of using 5 4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and then.

Standby. Uav online. Blaze. Double blaze. Triple glaze. Double kill. Viper. Wrong. Cop is up wins. Team death match. We are experts. Triple Blaze. Let's just roll, right? Market clearing order inbound. You're surrounded by Jungler. Hold your position. Strike 1. Strike 2. Activate. Go. Golden retriever mode. Clearing order in. Uav online. Five quarter. I see multiple journalists on the horizon. Stand by, Founder. You're watching TVPN. Today is Monday, March 16, 2026. We are live from the TVPN Ultradome. The Temple of Technology, the Fort Fortress Finance, the capital of capital. Let me tell you about ramp.com time is money save. Both easy as corporate cards, bill pay, accounting, and a whole lot more all in one place. What a massive week last week. Alex Karp going back to back with Travis Kalanick. The reactions to the Travis Kalanick interview was phenomenal. I was reading them all weekend. I was still emotional the next day. Yeah. And there's something I posted this on one of those clips that someone just shared was like, this is a great clip. And I was there because you're in the moment and you go, do I reflect too much on different interviews? There's always the next day of interviews. But you know, watching some of the clips back, Guillermo from Vercel put together that hour long, so good kind of motivational video. Yeah. It was so good. I think that the Travis Kalanick mindset has been missing totally. When he kind of left, there was. There's been a Travis sized hole. Yeah. In the industry, in the culture. Yeah. And to see him come back and in, you know, 45 minutes, basically just give the advice that I think, like everyone that's building in some way can benefit from. Not everyone is going to be Travis, but there isn't anybody out there that's done what Travis has done that is kind of like preaching that. And I don't like listening to like founder porn content. Personally, it's not appealing. But when it comes from Travis, it is just another level. Yeah. Like the right message at the right time. Yeah. Especially the thing, the thing that I was kind of pulling on is like right now, like, there's a lot of easy money everywhere. Right. There's teams that have built nothing that can raise between 50 to a billion dollars at times. And in his feedback on that, his point of view was like, okay, is capital really a constraint in your business? How much does it matter? How much is it going to matter in terms of the competitive dynamics of your market and if it matters? And in a lot of these AI categories, it does and if it matters and it was easy, that means you didn't go hard enough. Yeah, yeah, that was the best line. And that was like the best line, like, if money matters, as, as we all agree, you raised a billion, why didn't you ra. If money matters, why didn't you raise 3 billion? Oh, it was easy. That means you didn't go hard enough. Yeah. I mean, that's somewhat the subject of what Dylan Patel was talking to Dorkesh about on the Dorkesh Patel podcast. Fantastic show, by the way. Fantastic episode about this being risk, on being aggressive. And Ben Thompson wrote about that today through a different lens, talking about, are we in a bubble? Maybe. But all the numbers are penciling out, so go, go, go. Now is the time to. Yeah, it was fascinating hearing it from a completely different perspective at the perfect time. But I really. Yeah, that was a great, great interview. That was, that was personal highlight for sure. Building TVPN for sure. Friday. Yeah, that was great. It was, it really was like, like the conversation that we set out to have because, I mean, he mentioned he's leaving California, but we're not going to like get bogged down in like his political views or whatever like that. It's. It's so much more about the actual craft of scaling a business. And like, I think, I think we just nailed that and so that was really fun. Yeah. And the good thing is we have plans to do a show like that every single day of the year. Unfortunately, it's not possible. Right. It's not very often that someone like Travis, world historic founder, comes out of media retirement after almost a decade. Almost a decade. Yeah. So very special. But thank you for everyone who tuned in. Thanks everyone who enjoyed any of the clips, saw whatever you saw of it. It was a really fun time. And if you care if you want to work in physical AI. Yeah. And you don't see yourself in the Elon verse. Yeah. I think that is one of your best possible bets. Sort of like an indexed approach to work insanely hard for Elon, work extremely hard for Travis. Yeah. Or you're going to have to work hard one way or another. Exactly. But that's nature of what you. If you don't want to work hard, there's probably a company out there that's competing with Travis or Elon in physical AI. You could work there. I just wouldn't put much value on your. On your RSUs. Yeah, it's rough. Anyway, let's pull up the linear lineup. Linear, of course, is the system for modern software development. 70% of enterprises Workspaces on LINEAR are using agents. We have Kevin Espertu from Epic Gardening coming on to tell his story about scaling his YouTube channel. I think we have a lot to talk about. We always love creator economy stories. Paul Coiningham, the dog healer is coming to break down how he used AI to augment delay his dog's cancer. We're gonna be digging into. We'll first go through what actually happened. I have some opinions about this. And then we'll talk to him to get his side. Then we are pulling our delayed lightning round. We went far longer than we expected with Travis on Friday. So we are catching up on our lightning round with Tony from Sunday Robotics and we have drew from 8eight VC is a founding partner there. HEC team backed Quince at seed. It's now a $10 billion company. The Quince founder is a little under the radar totally. But I wanted to get this story from the 8VC team, hear how they're thinking about it and then a bunch of other folks joining. So looking forward to that. Well, let's read through Brandon Gorell's deep dive on the AI versus dog cancer. What happened so late Friday? There was a story about an Australian tech entreprene, Paul Coynningham, reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Cunningham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's great to hear that. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Cunningham used ChatGPT to brainstorm ways he could help. And he did an interview on this and here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing, who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark, Embark DNA. You could do it. Okay, well anyway, he went to a university, probably for a good reason, probably got good data. He said the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 kilometers down the road. You can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Coynningham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of like, how much was this cure my dog cancer one shot. It don't make mistakes. I don't think anyone's saying that. But very quickly there was like an incentive to amplify this into like the hype, like this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. So once the DNA sequence was produced, he ran it through a whole bunch of custom different data pipelines to find those mutations and then used other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA institute, which used Cunningham's data crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us. And we made a little nanoparticle and it's democratizing the whole process. They said, this is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Cunningham and his team administered the vaccine to Rosie, which was affected. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have like a little bit. The question is like, is it runaway? Is it bad, is it terrible? And then it's hard to just like snap your fingers. And cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X news of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog. Air horn for Rosie. The air horn for the dog. That's great. Turned into a heated debate on health regulation after biomedical engineer Patrick Heiser posted that quote, it is trivially easy, trivially, trivially easy to make a single MRNA vaccine. It's not hard. And Hank Green, prominent YouTuber issued something of a rebuttal, which we can go through later. A separate thread in the discourse is focused on the promise of LLMs democratizing access to medical science, with OpenAI President Greg Brockman quote tweeting the story with the caption, a small window into the opportunity of AGI. Well, Coiningham didn't literally cure Rosie's cancer with ChatGPT. As Stripe CEO Patrick Collison pointed out, it acted as a high powered search tool that ultimately helped his team get to an amazing outcome. Sort of. George Hobbes. We got to move the goalposts. I'm ready to move them, I think. Moving the goalposts. I mean, where are we moving them to? It has to actually. You have to be able to type cure my cancer. And then from your phone it just deposits a pill that you just take. Yeah, exactly. It has to. Locally. Yeah. End to end. No, ideally. Ideally it would be not even a pill that you take. It can just create a video that you. The right pattern of light. The right pattern of light coming from. And sound. So the phone has light and sound and so the light flashes in your eyes at a certain rate. It rewires your brain and your brain decides to go kill the cancer. Yeah, we've talked about this a bunch. I think it would be helpful for the industry to refocus some messaging on not AI is going to cure cancer, but human humanity is going to use AI to cure cancer and do a number of other things. Right. And so the, the bar is not just like one shotting it with a prompt and it sends it to a lab and you get a, you know, some, some type of treatment in the mail. Maybe we, we. I can imagine that in the future. Right. Something, something to that effect, but it is an enabler, it's a tool. And this has allowed someone to become not an expert in something, but. But to help somebody understand a process enough to go out and find the right experts to help them solve their problem. And I think it's incredibly inspiring. So excited to have him on the show later. So there was a chemist who works in AI and biotech by the name of Ash Jogalikar, and he had a really good summary along those lines with a riff on Freeman Dyson's 2007 New York Review of Books essay R Biotech Future, which we should read at some point in which, in this, in this article, Freeman Dyson argues that biotechnology will become small and domesticated rather than big and centralized. The. The full post is worth reading in full and we might go through it, but the conclusion is particularly good. If AI continues to reduce the cognitive overhead required to navigate biological knowledge and assemble complex pipelines, the boundary between professional research and motivated individuals may begin to blur. That shift will require careful thinking about safety, governance, and responsibility. But it also carries an exciting possibility. Dyson imagined a world in which biological design might eventually become something like a creative craft practiced not only by institutions, but also by curious individuals experimenting at smaller scales. Yeah, I think there's the reality of cancer treatment, from my understanding is, and this was based on a late family member that had cancer and ultimately passed away. During the process, during his treatment process, which was around a year and a half, he was getting looks at different treatments that were promising, some of which he was able to do, some he didn't qualify for just based on his personal situation, even though there was, there was a decent chance that it could have had a positive effect. Yeah. And that sort of the insane frustration that an individual feels or a family feels when they're like, hey, this, you know, if, if something's terminal or it's looking really bad, it's progressing the wrong direction. And there's a, there's a treatment out there that isn't. That is somewhat trivial to actually make. You just don't qualify for it. That level of frustration will eventually drive more individuals, I think, to do this right. And so there's definitely, definitely some like safety. There's huge safety concerns, there's ethical concerns. These are things that we have to work through. But ultimately I just think there's going to be so much, there's going to be enough like human energy and just overall desire to live that people will take risks that they wouldn't take for a bunch of other more sort of like trivial sort of issues. There have been initiatives with the fda, something around. Right. To try in certain scenarios, patients rights, sort of removing some of the regulation and allowing people to make decisions like that. It does feel like the FDA's stance might need to change in this case. Like they clearly have a role to play currently and in the future where biotech becomes more democratized. But yeah, hopefully there's some good symbiotic relationship there with the broader biotech community as we get bigger. I have a similar story with someone who developed a rare illness and was able to go and read academic research at a very deep level. Didn't have a background in biotech or anything like that, but was able, this was pre AI, was able to read like every published research paper that was at all related to this particular illness and found the world expert in this particular disease contacted the professor and the professor said, yes, you have the thing that I've been studying and I've only found five people or 10 people in my entire career that have this thing come down, I will operate on you. The operation happened, it was successful and it was fundamentally like a high agency person doing a lot of research. And if AI just acts as a search tool that democratizes that, you're going to get better results. So even if you're, even if we're not in like one shot in curing cancer, that just feels like making search easier, making research easier. Huge benefit. Yeah, the guy Australia could have done a lot of this 10 years ago. He just would have needed to spend, I'm sure a bunch of time in. Yeah, that's the thing. Libraries, everything you do, all these things can do manually. You can, you can just get a guy for that. Yeah, you can get a guy or you, I mean you don't even need a spreadsheet. You can do, you can do this, you can calculate the math by hand. But these things speed things up. So it's, it's been a good time. Let's read through Ash's post. But first let me tell you about public.com investing for those that take it seriously. Stocks, options, bonds, crypto, treasuries and more with great customer service. And let me also tell you about fin, the number one AI agent for customer service. If you want AI to handle your customer support, go to fin A thank you for clapping, Tyler. How was your weekend, Tyler? It was good. Yeah, it was good. Yeah. Did you go to any data centers or are you. No data centers this weekend? I was in sf. Didn't you go to a pig roast? Yeah, that was on Friday. I was in El Segundo. How was sf? Is something big happening there? Does it feel like being in Wuhan in February of 2020 something. Something big was happening. Yeah, there was. I went to a debate. Oh, you went to the debate. Okay, cool. How was that? It was good. Yeah, yeah. It was about the billionaire tax. Yeah, yeah, yeah, yeah. And did you go to the hackathon at all? No, I missed that. Okay. I saw that semi analysis had a hackathon. The winners were crowned. Seemed like a lot of fun. It really does seem like the best time to go to hackathon just because what you can actually accomplish in two days is remarkable. Yeah, yeah. No, right, yeah. People used to do tagathons and it'd be like after two days they'd be like, we have a landing page. And now it's like. And a cool idea. We created a hackathon simulator with mini games for everything and it's also making money. We need to give an update on TVPN simulator at some point, but it is coming along. The development has continued at breakneck pace. Yeah, we got to work on the rollout of this. Yes, it might be GTA 6 level by the time GTA 6 comes out. I think we can get there. We need a new graphics package. What do you think the actual path to AAA graphics is? Do you think we should rewrite it in Unreal Engine with ray tracing and insist that people only run it on gaming PCs or should we do some sort of style transfer on top of it? Yeah, I think the Unreal Engine is probably easier because you're just moving the code over. That shouldn't be like that difficult. You probably do that in like a day or two. I think these things used to take like years. Like it took. It took Elder Scrolls like a decade to get to like Nintendo Switch. Yeah, it takes a day or two. The real time render thing is interesting, but I think that's. It's just like expensive. That's the problem. We had. We had someone on the, on the show that was doing it on Zoom over in real time. That was Descartes. Descartes, yeah. That was a cool demo. Yeah. So you imagine like that tech prompted with like make. Take this from like boxy. I would say we're at. We're above N64 level graphics, but we're probably more like Xbox 360 graphics and take us into modern day PS4. Yeah. I mean this is why I'm very excited about doing. Everyone's so up in arms about like, oh, the new PSX isn't going to come out because you ain't the memory. And people are like, oh, I don't want to play Games in the cloud. Right. But if you're in the cloud, that means you can actually like access a ton of compute. Because like when you're not playing, when you're not using the GPU to run like the nice graphics, someone else can be. Yeah, you can actually get higher to. You can get access to much better like hardware playing video games and then also. Yeah, more, more iteration on the graphics. Like it should just be like live service model basically. Yeah. And if you get the, you know, the Genie 3 model where it's actually, you know, generating on spot, like. Yeah, that's something you can really only do in the cloud. I'm excited. Jensen is doing his keynote at gdc. Should we pull up the live stream? We can. Yeah, let's check in with Jensen. Let me tell you about Okta first. OKTA helps you assign every AI agent a trusted identity. So you get the power of AI without the risk. Secure every agent, secure any agent. And let me also tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Continue. Let's play it. What do we got? We got Jensen, institutional investor. These three people are deep in technology, deep in what's going on. And of course they have just a really broad reach of technology ecosystem. And then of course all of the VIPs that I hand selected to join us today. All star team. I want to thank all of you for that. All star team. The leather jacket really has just aged so well. I also want to thank all the companies that are here. Nvidia, as you know, is a platform company. Mic drop. We have technology, we have our. Oh, by the way, everyone uses. He's mogging our merch. He is. Today there are probably a hundred percent of the hundred trillion dollars of industry here. 450 companies sponsored this event. I want to thank you a thousand, I love it. Technical sessions. 2,000 speakers. This is 2,000 speakers. Wow. Every single layer in one. They're going to do more interviews than we've done all year. In one chips for two days to the platforms, the models and of course the most important and ultimately what's going to take get this industry taken off is all of the applications. This really is the super bowl for semiconductors. It all began here. This is the 20th anniversary of CUDA. We've been working on CUDA for 20 years. For 20 years we've been dedicated to this architecture, this revolutionary invention. Simt, single instruction, multi threading. All right, very, very cool. Let's get back to the timeline. Let's go to Gemini 3 Pro. Gemini 3.1 Pro is here with a more capable baseline. It's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life. And let me also tell you about Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases and more, while Railway automatically takes care of scaling, monitoring and security. Back on the timeline, we gotta go through. Wait. Yeah, you can just save your dog with a beautiful picture here. You can just save your dog. It's remarkable. This is a heartwarming story. And it also, yeah, I really like how it reveals current AI capabilities where things are the benefits and sort of the diffusion narrative. This is fundamentally a diffusion story, not a super intelligence story in my opinion. But let's go through Ash's post here. My take on the whole AI cures dog cancer in Australia is a very interesting story, but perhaps not for the reasons that are being noted. In 2007, Freeman Dyson published an essay in the New York Times Review of Books called Our Biotech Future. It contains one of the most memorable predictions about the future of biology that I've ever read. I predict that the domestication of biotechnology will dominate our lives during the next 50 years, at least as much as the domestication of computers has dominated our lives during the previous 50 years. Dyson believed biology would eventually follow the trajectory of computing. At first, powerful tools live inside large institutions, universities, government labs, major companies. Over time, these tools get cheaper, easier to use, and more widely distributed. Eventually, individuals start doing things that once required entire organizations. You will be the manager of infinite minds. You will have, you know, a million agents, and you will also have access to the equivalent of a university lab filled with biotechnology equipment. Biotech will become small and domesticated, rather than big and centralized. This is very interesting in the age of AI because there's been this narrative of like, AI is a centralizing technology, it is very power law driven. But this is sort of counter to that. I don't exactly know how to piece those two things together, but it is interesting that his prediction was actual decentralization in this particular category. He even imagined genome design becoming almost artistic. Designing genomes will be a personal thing, a new art form as creative as painting or sculpture. Dyson's words rang in my mind as I read the AI cures dog cancer story. Much of the coverage framed in. I gotta say, it's very easy to imagine you in 20 years. I'm like John like, you gotta tell us your anabolic steroids. And you're like, it's kind of a personal thing. It's kind of a personal thing. It's kind of like an artisanal process that I go through the sculpture. I'm sort of sculpting myself. I can't really. I'm sorry, I can't really share my stack with you, but it's a personal thing, so go and kind of figure out your own stack. You know, speaking of sculptures, I was walking around my neighborhood and I looked through this, like, you know, gap in the trees into this, like, large lawn. And I saw on this person's like, front lawn, behind, like, you know, gates and whatnot, just a full size statue of a man playing golf who I didn't recognize. It was like, it was not tiger wolves. I think it was the owner. I think it was the owner. I think the owner was like, I'm into golf. Or one of his boys got it for him, which is a hilarious gift. Getting someone a life size statue of themselves and just having it delivered. And then it's like, well, it's impolite for you to turn it down. What are you gonna do? Anyway, the scientific pipeline involved here is actually well known. It closely mirrors the workflow used in personalized neoantigen vaccine research that has been under active development for years. The steps are fairly standard. Sequence the tumor, identify somatic mutations, predict which mutated peptides might be recognized by the immune system, encode those sequences into an MRNA construct and deliver them to stimulate an immune response. The biological targets themselves were almost certainly not new discoveries. I have been unable to find out what they are. But mutations in targets like kit, which are common, might be involved partly. Therein lies the rub, since the hardest part of drug discovery, whether in humans or dogs, is target validation, the lack of which leads to a lack of efficacy. The number one reason for drug failure. In neoantigen vaccines. The proteins involved are usually ordinary cellular proteins that happen to contain. Contain tumor specific mutations. Alphafold, which was used to map the mutations onto specific protein structures, is now a standard part of drug discovery pipelines. That's fascinating. The challenge is identifying which mutated peptides might plausibly trigger immunity. What is interesting though, is how the pipeline was assembled. Normally, this type of workflow spans multiple domains. Genomics, bioinformatics, immunology and translational medicine. And in institutional settings, those pieces are distributed across specialized teams, document sources, and legal and technical barriers. Navigating the literature, selecting computational tools, interpreting sequencing results, and designing a candidate MRNA construct is typically a collaborative process. In this case, AI appears to have helped compress that process, pulling together data and tools from different sources. Instead of requiring multiple experts, a motivated individual was able to assemble the workflow, with AI acting as a kind of guide through the technical landscape. That is fascinating. Anyway, it's a longer post, but you should go read the thing in full. Patrick Collison also chimed in. He said, according to the story, the dog's cancer has not been cured. I think it's just 50% smaller, which of course is a win. But not using the term cure is always tricky, but it does go viral. Absent all regulatory and manufacturing constraints, we could not just synthesize magic RNA, RRNA cancer cures. The technology is very promising, but it's not any kind of panacea. Yet the emergent system, emergent system of system of regulators and manufacturers is indeed far too conservative and small. Small scale experimentation is much harder than it should be. More people should read the first part of the Raw the Rise and Fall of Modern Medicine. So it's interesting, Lee says chatgpt Cure cancer, Make no mistakes Biomedical engineering industry yeah, don't do this. It's easy and effective, but we can't make enough money off of it. Ridiculous. It's surprising. G. Fodor says it's surprising how people are so blatantly talking past each other on this. The point is that the system of clinical trials is predicated on an assumption that a given drug will work on a cohort. What if there are lots of drugs that will only work on one person? So. So definitely a big desire and push for rethinking the system of clinical trials. If you're going to have personalized medicine, what does that mean? There's already a lot of people, biographers that do all sorts of stuff like this. Can't believe he wasted two cups of water to do this. Banai it is ridiculous. It's a great counterpoint to the doomers. What else is going on? Marc Andreessen chimes in I can't load the post right now. We gotta go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32 inch MacBook called the MacBook Pro Ultra Wide and it looked like this. I bought one and unlocked extreme productivity and then it wouldn't fit into my backpack so I had to leave it behind. Oh no. This is sort of like on that other laptop that was they should honestly make this. They should. I mean, walking around looking like maybe you could put skateboard Trucks on it. Yeah. That you could use it as transport. Yeah, it's more of like a snowboard build that you like carry over your shoulder like this. Or surfboard. You know, people throw it on the top of your car like that, three fingers. Why you don't put a surfboard on the top of your car? Yeah, I mean, real ones don't. Oh, what do they do? They put it inside the car truck bed or inside the car truck bed. Okay. Yeah. I don't pretend in the LA area, you can clock if somebody's actually a surfer or not. Just by the way they go. Each with their board. Okay. No, but I think. Yeah. Throwing it under your arm, having some trucks on it, skating, being able to get where you need to go. I like the ultra wide. What if they're driving a Huracan sterrato? Where would you recommend that they put their surfboard then? Jordy Stirato. I can make exceptions. Okay. I like this. Dylan Patel said on Dwarkesh, The TAM for GPT 5.4 is north of $100 billion. But there's adoption lag. That's considered AGI as far as the Microsoft OpenAI contract is concerned. That's very interesting. Sam Carter says the reported 1 billion of profit is no longer the sole trigger for confidential IP research access. It reportedly includes an independent expert review. You were saying Joe Rogan would be on that, Andrew Huberman, the experts would be on there. You got to trust them at all times. And Neo Vaughn, maybe. You know, the funniest thing about that joke is that, like, I actually would like to know that panel of experts whether, where they deem AGI, because I feel like between all of them, they could have, they could chat with the chatbots and be like, ah, it's like not that good yet. You know, like, be very realistic about it. Yeah, they're not necessarily just going to be like, oh, I'm pumping it for whatever reason. They're like, I have this weird bias or whatever. And so it'll be very interesting to see how that, how, how the AGI definition plays out, because it does feel like we're close. I mean, Dario on Dorkesh was saying, like, we're near the end of the exponential, which is like, sort of crazy. It feels like, you know, Sequoia declared AGI, they're an investor in Open. And so there's a lot of stuff. What do you think about the AGI? Do you want to be on the expert plant panel? I think I would say that we already reached AGI. It was maybe earlier you called it like 30 seconds before Tyler Cowan did. I think I remember it was like 30 seconds before you came. You tapped me on the shoulder and you said it's here. And then we went and refreshed X and Tyler Cowan had come out. Yeah. I think realistically I'd probably say it was something like when the agentic harnesses came out. So stuff like Claude code. Sure. Where you can actually just tell it to build a project and then there'll be errors. It'll see those errors, it'll fix them. It'll like keep working on it. Not reasoning models. I mean it's so hard. It's like unlike math or something like this. Right. But those basically unlocked. Yeah. Now they can just do anything. Yeah, yeah. I mean the agentic thing was talked about for a full year and then it finally happened like in December and it was pretty broken up until then. And then. Yeah, but I think you can still just make a very good case that like. Yeah, chatgpt like that was AGI like you can just ask a question, it'll answer it. Yeah. If you'd never talked to an AI model before and you talked to that, you're like this. Okay, this is a person. Microsoft Excel 1985 AGI Jose Macedo says ultimate narrative violation from the Dylan Patel to our Keshe pot 3 years later H1 hundreds are actually trading above launch price in secondary markets I. E. Negative depreciation. Yeah, that's called appreciation. Appreciation. I just want to go out and say I appreciate age 100 negative depreciation. This completely flips the Michael Burry 2 year E waste bear thesis on its head. Yeah, I mean somebody's got to check on. Somebody's got to check on Michael Burry. It's such a different. It's such a different dynamic because I mean like the whole. There was a reasonable underpinning for GPU depreciation, which was just look at 20 years of computer equipment history. It's like it all depreciates over like maybe five years, maybe 10 years. Some stuff sticks around but like they burn. Yeah. It's just interesting. Jose says core, we probably benefits most from this. They have 250,000 GPUs and a $66 billion backlog. Depending where you think market was pricing depreciation margins improved by something around 40%, which means 1 billion a year in additional earnings. Who knows where this stuff actually re rates or. Or how sustainable. But great, great time to be a neolab 1 of 1 of the founding team members at Lambda Was posting last week. Basically congratulations to everybody that booked out like GPUs on, on. On an annual basis in 2025. You're looking like absolutely brilliant right now. Obviously Sam is starting to look extremely vindicated on all the deals that he did last year. So, Tomas, quickly let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB? Don't just build AI, own the data platform that powers it. And let me also tell you about TurboPuffer, serverless vector and full text search. Built from first principles and object storage. Fast 10x cheaper and extremely scalable. We've been growing a lot and are out of GPUs. This is Sam Altman in March of 2025. Satya Katz over at Oracle says we are still waiving off customers or scheduling them out in the future. This is a situation that we have not seen in our history. Satya says you may actually have a bunch of chips sitting in inventory that I can't plug in. I don't have warm shells to plug into. Sundar says, what keeps us up at night? The top question is definitely around capacity. All constraints, be it power, land, supply constraints, how do you ramp up to meet this extraordinary demand? And sorry, quickly, between power, land, supply chain constraints, chips. You were saying that the takeaway. Your takeaway or your read on Dylan Patel on Dwarkash was that chips were the main. Yeah. I mean, not even like a read. Like he explicitly said. He's like, between power and chips. Chips. Chips is what's going to be the big bottleneck. Because at some point, like. Yeah. There's all these ways that you can actually like, maybe get like 10% of the, you know, US energy production to just like go to. Yeah. You know. Yeah. Where like, at some point, like, okay, we don't have enough, like, EUV tools. Yeah. And like they're not building them right now, which means that they're not going to have them for at least three, four years. Yeah, yeah, yeah. This was Ben Thompson's. Like, TSMC needs to step up and spend more on CapEx. Their. Their CapEx guide is like a CapEx guide for ants. Like a mere 45 billion or something. And it should be probably much, much higher. Yeah. But I mean, it even goes down to the, you know, the toolmakers below them, like really, like really deep in the supply chain. Yeah. At least what I got from do tell on that interview is that like, they still are not really that AGI pilled. They're not expecting this Kind of massive, you know, increase in demand to stick around. Yeah. Trey says a sign of taste is dabbling in the vintage GPU market. A1 hundreds. A1 hundreds? Yeah. Ampere v. You gotta go Volta. Go back to Volta. Yeah, I mean I remember I was digging into that like chips versus energy. What's the big bottleneck? And I think we're using something like 50% of leading edge capacity like of the fabs that can make AI powered GPUs like GPUs that can run transformer based large language models. We're using like 50% of that capacity already. And then some of the leading edge nodes go towards like you know, Apple silicon chips that are maybe designed system on chip, something for a phone. And only like 1% of energy right now in America goes towards AI and or less it's like 0.1 or something. So you can reallocate and everyone just turn off your air conditioning. One more close the door. If the air conditioning lip Bhutan says there no there's no relief as far as I know. No relief until 2028. Somewhat ominous. Keep reading what Tomas says. What happens when your AI doesn't answer? Everything is in short supply. It's no longer just GPUs, it's power, data centers, memory, CPUs. If there's no relief for six more quarters, perhaps it's time to plan for a world where inference isn't freely available on demand. Inference prices which have been static will rise. Subsidies will be harder to justify. Enterprises will need to rationalize workloads deciding which teams receive state of the art models and which don't. Not every CRM update requires a trillion parameter frontier model. Inference rationing normalizes. Market marketing receives this much sales receives that much. Software engineers probably receive a lot more Constraint will be the mother of invention. Companies will optimize what they have, adopt open source where they can and likely move to smaller models for many workloads. Really cool take. I like this. It's also interesting to me is that not every CRM update requires a trillion parameter frontier model. That's another bull case for Hopper price stability going forward and less depreciation. Because if you can leave your CRM update workload where you're just going through and spell checking names and cross referencing data sources and pulling from email and dumping some notes, summarizing. And you can do that on a GPT4 class model instead of using 5 4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently and so that's still economically valuable and so you're able to continue that. Should we go over Ben Thompson's post from this morning? Yeah, we should now be a good time. Yeah. First, let me tell you about Label Box, RL Environments, Voice robotics, evals and expert human data. Label Box is the data factory behind the world's leading AI teams. And let me tell you about Vibe Co, where D2C brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audience, measure sales. Just like on Meta Agents published this this morning. To me, the second I saw that I started reading it. It felt like taking a double scoop of C4. Is that a pre workout? Yeah, I know the can. I didn't know it was a. You never, you never dabbled. What was the one that we. I'm more of the gorilla mind one. That's the one that I. Many people have said you have the mind of a gorilla. Yes, yes, yes. For more plates, more dates, you're a gorilla in sheep's clothing. I think that's literally the pre workout that I have, although I don't use it often anyway. So you got pumped up? I got pumped up. Ben writes, there's a weird paradox in terms of AI prognosticization. Prognostication. Prognostication. That was a good, good effort, Jordy. On one hand, what are some of these requirements? There's just so many words. What are the requirements for having a podcast? Like knowing how to say words. No taste. I mean, yeah, ultimately there's a lot of words that you, when you read them. Yeah, you're just like, oh yeah, you can just do it. And then you try to rip it. On one hand, you don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out? To be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I very I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026. However, on the morning of Nvidia GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which, paradoxically, may be the truest evidence we are. Where's the bubble gun? Let's get the bubble gum going. He writes. LLM paradigms over the last couple of weeks. First in the context of Nvidia earnings and then last week in the context context of Oracles. I've talked to you about 3lLM inflection. I've talked about 3lM inflection points. I'm not going to go through all of these. He goes chat. We've talked about this a few times. LLMs, reasoning models and then agents. And each one of those increases the demand exponentially for compute. Yeah, so LM, ChatGPT01 and then Opus as well as Claude Code and Codex. Codex basically getting the point where tasks are being accomplished over hours and getting to great outcomes. And this is the interesting point, okay, the decreased need for agency the reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the capex by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the model so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of capex expenditure not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that compute and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they are going to be used even more than even reasoning models in a chat bot. It's how this third point will be manifested that I think is underappreciated. After all, far more people use chatbots than agents. But I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to use AI. And he goes into a little bit talking about local local compute, talking about how Apple's opportunity to run LLMs locally. There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599 I think 499 for education. Potentially very disruptive to other laptop makers. You still get discounts, Tyler. Or does it? I think I'm still scared. Oh, yeah, you're still. Because you're on leave. That's great. There you go. There are some legendary leave of absences where people have been away for like 10 years and then they go and do so many. See, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before, while you're still eligible. That's a good one. Because they're like, oh, well, we gotta. I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year and I think they gave him the honorary degree a couple years later. So, you know, that's the speedrun to beat. But the point about the MacBook Neo is that at $599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with like Asus or another brand? I'm not in the Apple category. Like, it's not an option because that store over there, those, those laptops start over 1,000. That's not my budget. So I'm not even going in that store. Well, now you can. And you can spend $600 and get a pretty good computer. And the CFO Nick Wu of Asus was on their recent earnings call and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments in the second half of last year. We made some internal prep, but now that it's out, we don't think it's that big of a deal. Like, it has some limitations. Specifically, it only has eight gigs of ram. So, like, you know, this is more focused on content consumption. It's not a mainstream notebook for notebook usage, for creation, for working. It's not a work device. It's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, that's what people use these laptops for now. It is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more and then he goes on to apply that to. To AI, talking about enterprise and the value of. Companies have a demonstrated willingness to pay for software that makes their employees more productive. And AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the pro, not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, basically my interpretation, he's making a case that there are companies that could cut headcount and actually just grow faster. Yeah, if they're implementing AI properly, not just replacing like the routine workloads. Yep, so he says. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay, actually, I'm going to start one paragraph. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through huge apparatus filled with humans who accelerate the effort in some vectors and retard it in others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard to manage and motivate human cogs in the organizational machine. With agents that not only do what they are told, but do so tirelessly and continuously until the job is done. This only makes the argument that we are not in a bubble much more compelling. Unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously all. I'll get around to it when I feel like it, I'll give it a crack. I'll get around. Yeah. So this only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of alums are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing. Right. You have one. Tyler just, you know, he's going to. Tyler's going to set up to be able to do sign language with his agent to just be. Not even speaking, just sending. Did you actually ever use any of the voice models? Remember Carpathy was talking about that? Yeah, a lot of people do this because you can just talk much faster. I guess I Haven't done this really. I've used it sometimes use the voice mode, but I don't actually use it in coding agents yet. I was using the ChatGPT voice mode. Like the true back and forth voice mode? Yeah, like real time voice. Real time voice mode in the car this morning. And they improved that thing dramatically. It's good. Yeah, it's so much better. So first off, it doesn't do that like that's a great question or anything like that, or that whole pause that was in the super bowl ad that just doesn't exist anymore. It just answers and it answers in these really short, punchy things. I was asking it about how many jobs are actually in America and it just says like 164 million. And it just like gets me the answer. And I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more detailed back and forth without needing to like dictate a whole prompt and then let pro cook on it for 10 minutes, come back, have it read it. To me, it was like a much better experience. I was very pleasantly surprised by how the back and forth worked. And they also changed it so that you see the floating bubble of like the little animation, but the text populates in real time with your question and then the answer and then your question and the answer. So you can just scroll and read as well. It's very cool. Anyway, third, the last argument that we are not in the bubble. The economic returns from using agents aren't just impactful on the bottom line, that is saving on cost, but the top line as well. Let's go in this context. It is any wonder that every single hyperscaler says the demand for compute exceeds supply and that every single hyperscaler is in the face of stock market skepticism, announcing CapEx plans that blow away expectations. So I encourage you to go subscribe to strategy, max out your plan, pay annual. But this was extremely notable. It's such a funny ending where he, he has this point about like, you only need to be worried about a bubble, when like, you don't need to be worried about a bubble if everyone's saying a bubble. Because then everyone's like, risk off because everyone agrees that, oh, we're in a bubble, let's not do bubble behavior. And so capitulation is the sign of a bubble. And he's like, I understand that. And still this is my take. It's a bold take, but I think It's a good one. Really quickly. Let me tell you about the New York Stock Exchange. Want to change the world, Raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets. Take us through it, Jordan. He was going hard. He was going hard. Yeah. We'll click into this top Apollo executive sounds off on arrogance in private markets. You always want to be sounds, he says. I literally think all the marks are wrong. Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives of the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co president of the firm's asset management arm that is one the of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or mid sized Joe software company might recover 20 to 40 cents on the dollar, and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary, according to audio recordings of the comments reviewed by the Wall Street Journal. It sounds like this wasn't meant to be public. I don't know. Calling people in private markets arrogant is crazy. I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold call by him, but we'll see what evidence he has to back up that extraordinary claim. He blamed the media for creating a frenzy around privacy credit. Obviously we're in the middle of a private credit party. Apparently if you do credit well, it's honestly, I would say we don't understand private credit well enough to like really put, put everyone up into a frenzy, he says. If you do credit well, it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. Zito talked about the sell off in shares of large software companies, which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity, many with private credit loans, could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good so all is good. Anyone who says that clearly doesn't understand. Most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public yet smaller than those companies and we're trading at a much higher valuation than those companies. And so I am concerned about many of those. Take privates. Yeah, I remember a lot of like Logan Bartlett was doing a ton of analysis in the end of the ZIRP era at how high the multiples were in the public markets and that's what was driving the 100x ARR transactions. And you have to imagine that even if we were like oh yeah, that VC backed company was sort of over hyped at 100x arrangements. Well that still has a trickle down effect to you know, the private equity buyout that's just like yeah. Remember last year when Figma went out? Yeah. And they priced it. Yeah. Very reasonably. Right. They were very intelligent how they priced it. But then obviously there was so much excitement because it's such a great company. Round trip ran up when it. In the first couple days there was. There were some late stage private SaaS companies that I remember were posting like maybe I should go to the. I think it was the Parker rippling was like oh if I can get. That's a crazy multiple. Yeah, if I can get some insane revenue multiple maybe this appealing. Obviously a lot of those names still could get out this year. Strong companies. They're not as eager to get out. Zito pointed To Thoma Bravo's 2021 6.4 billion take private of the software firm Medallia in particular several lenders to Medallia including Apollo have already written down its its debt. He says there will be an issue with respect to that credit which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic smaller mid sized Joe Software company Zito said Joe Software company if he's in the wrong place I think is going to recover somewhere between.

Is justified. The first paradigm required a lot of compute for training but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the model so much more useful meant that they were used more which drove increased token usage in its own right. It's a third paradigm however that has truly tipped the scales in favor of capex expenditure not being speculative investment but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute and that compute and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness which means they are going to be used even more.

Is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens, because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the model so much more useful meant that they were used more, which drove increased token usage in its own right. It's the third paradigm, however, that has truly tipped the scales in favor of capex expenditure not being speculative investment, but rather badly needed and.

Paul Coyningham, the dog healer, is coming to break down how he used AI to augment delay his dog's cancer. We're gonna be digging into. We'll first go through what actually happened. I have some opinions about this. And then we'll talk to him to get his side. Then we are pulling our delayed lightning round. We went far longer than we expected with Travis on Friday, so we are catching up on our lightning round with Tony from Sunday Robotics. Then we have drew from 8VC. He's a founding partner there. He 8VC team backed Quince at seed. It's now a $10 billion company. The Quince founder is a little under the radar. Totally. But I wanted to get this story from the 8VC team, hear how they're thinking about it, and then a bunch of other folks joining. So looking forward to that. Well, let's read through Brandon Gorell's deep dive on the AI versus dog cancer. What happened so late Friday? There was a story about an Australian tech entrepreneur named Paul Coynningham reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Cunningham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's true. It's great to hear that. Australia, after his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Queen Ham used ChatGPT to brainstorm ways he could help. And he did an interview on this and here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark, embark DNA. You could do it. Okay. Anyway, he went to a university, probably for a good reason, probably got good data. He said, the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 km down the road, you can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Coynham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of like, how much was this cure my dog cancer, one shot, it don't make mistakes. I don't think anyone's saying that. But very quickly there was like an incentive to, to amplify this into the hype, this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. So once the DNA sequence was produced, he ran it through a whole bunch of custom different data pipelines to find those mutations and then used other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA institute, which used Cunningham's data, crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us. And we made a little nanoparticle and it's democratizing the whole process. They said, this is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Corningham and his team administered the vaccine to Rosie, which was affected. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have like a little bit. The question is like, is it runaway, is it bad, is it terrible? And then it's hard to just snap your fingers and cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X, the News of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog. Air horn for Rosie. The air horn for the dog. That's great. Turning to a heated debate on health regulation after biomedical engineer Patrick Heiser posted that quote, it is trivially easy, trivially, trivially easy to make a single MRNA vaccine. It's not hard. And Hank Green, a prominent YouTuber, issued something of a rebuttal, which we can go through later. A separate thread in the discourse is focused on the promise of LLMs democratizing access to medical science, with OpenAI President Greg Brockman quote, tweeting the story with the caption, a small window into the opportunity of AGI. Well, Coiningham didn't literally cure Rosie's cancer with ChatGPT. As Stripe CEO Patrick Collison pointed out, it acted as a high powered search tool that ultimately helped his team get to an amazing outcome. Sort of. George Hobbs. We gotta move the goalposts. I'm ready to move them, I think. Moving the goalposts. I mean, where are we moving them to? It has to actually. You have to be able to type cure my cancer. And then from your phone it just deposits a pill that you just take. Yeah, exactly. Is that what it is? It has to. Locally. Yeah. End to end. No, ideally. Ideally it would be not even a pill that you take. It can just create a video that you write. Pattern of light. The right pattern of light coming from. And sound. So the phone has light and sound and so the light flashes in your eyes at a certain rate. It rewires your brain and your brain decides to go kill the cancer. Yeah. And we've talked about this a bunch. I think, I think it would be helpful for the industry to refocus some messaging on not AI is going to cure cancer, but human humanity is going to use AI to cure cancer and do a number of other things. Right. And so the bar is not just like one shotting it with a prompt and it sends it to a lab and you get some type of treatment in the mail. Maybe I can imagine that in the future. Right. Something to that effect. But it is an enabler, it's a tool. And this has allowed someone to become, not an expert in something, but to help somebody understand a process enough to. To go out and find the right experts to help them solve their problem. And I think it's incredibly inspiring. So excited to have him on the show later. So there was a chemist who works in AI and biotech by the name of Ash Jogalikar and he had a really good summary along those lines with a riff on Freeman Dyson's 2007 New York Review of Books essay our Biotech Future, which we should read, I think, some point in which in this article, Freeman Dyson argues that biotechnology will become small and domesticated rather than big and centralized. The full post is worth reading in full, and we might go through it, but the conclusion is particularly good. If AI continues to reduce the cognitive overhead required to navigate biological knowledge and assemble complex pipelines, the boundary between professional research and motivated individuals may begin to blur. That shift will require careful thinking about safety, governance and responsibility. But it also carries an exciting possibility. Dyson imagined a world in which biological design might eventually become something like a creative craft practiced not only by institutions, but also by curious individuals experimenting at smaller scales. Yeah, I think there's the reality of cancer treatment from my understanding is, and this was based on a late family member that had cancer and ultimately passed away. During the process, during his treatment process, which was around a year and a half, he was getting looks at different treatments that were promising, some of which he was able to do, some he didn't qualify for for just based on his personal situation, even though there was a decent chance that it could have had a positive effect. And that sort of the insane frustration that an individual feels or a family feels when they're like, hey, this, you know, if something's terminal or it's looking really bad, it's progressing the wrong direction and there's a treatment out there that isn't, that is somewhat trivial to actually make. You just don't qualify for it. That level of frustration will eventually drive more individuals, I think, to do this right. And so there's definitely, definitely some like safety. There's huge safety concerns, there's ethical concerns. These are things that we have to work through. But ultimately I just think there's going to be so much, there's going to be enough like human energy and just overall desire to live that people will take risks that they wouldn't take for a bunch of other more sort of like trivial sort of issues. There have been initiatives with the FDA, something around right. To try in certain scenarios, patients rights sort of removing some of the, some of the regulation and allowing people to make the make decisions like that. It does, it does feel like the FDA's stance might need to change in this case. Like they clearly have a role to play currently and in the future where biotech becomes more democratized. But yeah, hopefully there's some good symbiotic relationship there with the broader biotech community as we get bigger. I have a similar story with someone who developed a rare illness and was able to go and read academic research at a very deep level. Didn't have a background in biotech or anything like that, but was able. This was pre. I was able to read like every published research paper that was at all related to this particular illness and found the world expert in this particular disease contacted the professor and the professor said, yes, you have the thing that I've been studying and I've only found five people or 10 people in my entire career that have this thing come down, I will operate on you. The operation happened, it was successful and it was fundamentally like a high agency person doing a lot of research. And if AI just acts as a search tool that democratizes that, you're going to get better results. So even if you're. Even if we're not in like one shot in curing cancer, that just feels like making search easier, making research easier. Huge benefit. Yeah. Cunningham, the guy Australia could have done a lot of this 10 years ago. He just would have needed to spend, I'm sure, a bunch of time in. Yeah, that's the thing. Libraries, everything. You can do all these things you can do manually. You can just get a guy for that. Yeah, you can get a guy or you. I mean, you don't even need a spreadsheet. You can do this, you can calculate the math by hand. But these things speed things up. So it's been a good time. Let's read through Ash's post. But first let me tell you about public.com investing for those that take it seriously. Stocks, options, bonds, crypto, treasuries and more with great customer service. And let me also tell you about fin, the number one AI agent for customer service. If you want AI to handle your customer support, go to fin. A. Thank you for clapping. Tyler. How was your weekend, Tyler? It was good. Yeah, it was good. Yeah. Did you go to any data centers or. No data centers this weekend? I was in sf. Didn't you go to a pig roast? Yeah, that was on Friday. I was in El Segundo. How was sf? Is something big happening there? Does it feel like being in Wuhan and February of 2020? Something. Something big was happening. Yeah, there was. I went to a debate. Oh, you went to the debate. Okay, cool. How was that? It was good. Yeah. Yeah. It was about the billionaire tax. Yeah, yeah, yeah, yeah. And did you go to the hackathon at all? No, I missed that. Okay. I Saw that semianalysis, had a hackathon, the winners were crowned. Seemed like a lot of fun. It really does seem like the best time to go to a hackathon just because what you can actually accomplish in two days is remarkable. Yeah, yeah. No, right, yeah. People used to do hackathons and it'd be like after two days they'd be like, we have a landing page. And now it's like. And a cool idea. We created a hackathon simulator with mini games for everything and it's also making money. We need to give an update on TVPN simulator at some point, but it is coming along. The development has continued at breakneck pace. Yeah, we got to work on the rollout of this. Yes, it might be GTA 6 level by the time GTA 6 comes out. I think we can get there. We need a new graphics package. What do you think the actual path to AAA graphics is? Do you think we should rewrite it in Unreal Engine with ray tracing and insist that people only run it on gaming PCs or should we do some sort of style transfer on top of it? Yeah, I think the Unreal Engine is probably easier because you're just moving the code over. That shouldn't be that difficult. You can probably do that in a day or two. These things used to take years. It took Elder Scrolls a decade to get to Nintendo sw. Yeah, it takes a day or two. The real time render thing is interesting, but I think that's. It's just like expensive. That's the problem. We had someone on the show that was doing it on Zoom over in real time. That was Descartes. Descartes, yeah. That was a cool demo. Yeah. So you imagine like that tech prompted with like make. Take this from like boxy. I would say we're at. We're above N64 level graphics, but we're probably more like Xbox 360 graphics and take us into, you know, modern day PS4. Yeah, I mean this is why I'm very excited about doing. Everyone's so up in arms about like, oh, the new like PSX isn't going to come out because ain't the memory. And people are like, oh, I don't want to play games in the cloud. Right. But if you're in the cloud that means you can actually like access a ton of compute because like when you're not playing, when you're not using the GPU to run like the nice graphics, someone else can be. Yeah, you can actually get hired to. You can get access to much better. Like Hardware playing video games. And then also yeah, more, more iteration on the graphics. Like it should just be like live service model basically. Yeah. And if you get the, you know, the Genie 3 model where it's actually, you know, generating on the spot, like yeah, that, that's something you can really only do in the cloud. I'm excited. Jensen is doing his keynote at gdc. Should we pull up the live stream? We can. Yeah, let's, let's check in with Jensen. Let me tell you about Okta first. OKTA helps you assign every AI agent a trusted identity. So you get the power of AI without the risk secur every agent secure any agent. And let me also tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Continue. Let's play it. What do we got? We got Jensen, institutional investor. These three people are deep in technology, deep in what's going on. And of course they have just a really broad reach of technology ecosystem. And then of course all of the VIPs that I hand selected to join us today. All star team. I want to thank all of you for that. All star team. The leather jacket really has just aged so well. I also want to thank all the companies that are here. Nvidia, as you know, is a platform company. Mic Drop. We have technology, we have our platform. Oh, by the way, everyone uses. He's mogging our merch. He is. And today There are probably 100% of the $100 trillion of industry here. 450 companies sponsored this event. I want to thank you A thousand. I love it. Technical sessions. 2,000 speakers. This is 2,000 speakers. Wow. Every single layer in one. They're going to do more interviews than we've done all year. In one, the chips for two days to the platforms, the models and of course the most important and ultimately what's going to take get this industry taken off is all of the applications. This really is the super bowl for semiconductors. It all began here. This is the 20th anniversary of CUDA. We've been working on CUDA for 20 years. For 20 years we've been dedicated to this architecture, this revolutionary invention. Simt single instruction, multi threading. All right, very, very cool. Let's get back to the timeline. Let's go to Gemini 3 Pro. Gemini 3.1 Pro is here with a more capable baseline. It's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life. And let me also tell you About Railway Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases and more, while Railway automatically takes care of scaling, monitoring and security Back on the time we got to go through. Wait, you can just save your dog. It's beautiful. The beautiful picture here. You can just save your dog. It's remarkable. This is a heartwarming story and it also, yeah, I really like how it reveals like, like current AI capabilities where things are the benefits and, and sort of the diffusion narrative like this is. This is fundamentally a diffusion story, not a super intelligence story in my opinion. But let's go through Ash's post here. My take on the whole AI cures dog cancer in Australia is a very interesting story, but perhaps not for the reasons that are being noted. In 2007, Freeman Dyson published an essay in the New York Review of Books called Our Biotech Future. It contains one of the most memorable predictions about the future of biology that I've ever read. I predict that the domestication of biotechnology will dominate our lives during the next 50 years, at least as much as the domestication of computers has dominated our lives during the previous 50 years. Dyson believed biology would eventually follow the trajectory of computing. At first, powerful tools live inside large institutions, universities, government labs, major companies. Over time, these tools get cheaper, easier to use, and more widely distributed. Eventually, individuals start doing things that once required entire organizations. You will be the manager of infinite minds. You will have a million agents and you will also have access to the equivalent of a university lab filled with biotechnology equipment. Biotech will become small and domesticated rather than big and centralized. This is very interesting in the age of AI because there's been this narrative of like, AI is a centralizing technology. It is very power law driven. But this is sort of counter to that. I don't exactly know how to piece those two things together, but it is interesting that his prediction was actual decentralization in this particular category. He even imagined genome design becoming almost artistic. Designing genomes will be a personal thing, a new art form as creative as painting or sculpture. Dyson's words rang in my mind as I read the AI cures dog cancer story. Much of the coverage framed in. I got to say, it's very easy to imagine you in 20 years. I'm like, John, like, you got to tell us your anabolic steroids. And you're like, it's kind of a personal thing. Kind of a personal thing. It's kind of like an artisanal process that I go through, a sculpture. I'm sort of sculpting myself. I can't really, I'm sorry, I can't really share my stack with you, but it's a personal thing, so go, go and kind of figure out your own stack. You know, speaking of sculptures, I was walking around my neighborhood and I looked through this like, you know, gap in the trees into this like large lawn and I saw on this person's like front, front lawn, behind like, you know, gates and whatnot, just a full size statue of a man playing golf who I didn't recognize. It was like, it was not tiger Wolf. You think it was the owner? I think it was the owner. I think the owner was like, I'm into golf or one of his boys got it for him, which is a hilarious gift. Getting someone a life size statue of themselves and just having it delivered and then it's like, well, it's impolite for you to turn it down. What are you going to do? Anyway, the scientific pipeline involved here is actually well known. It closely mirrors the workflow used in personalized neoantigen vaccine research that has been under active development for years. The steps are fairly standard sequence the tumor, identify somatic mutations, predict which mutated peptides might be recognized by the immune system, encode those sequences into an MRNA construct and deliver them to stimulate an immune response. The biological targets themselves were almost certainly not new discoveries. I have been able to. I haven't not. I have been unable to find out what they are. But mutations in targets like kit, which are common, might be involved Partly. Therein lies the rub, since the hardest part of drug discovery, whether in humans or dogs, is target validation, the lack of which leads to a lack of efficacy. The number one reason for drug failure in neoantigen vaccines, the proteins involved are usually ordinary cellular proteins that happen to contain tumor specific mutations. Alphafold, which was used to map the mutations onto specific protein structures, is now a standard part of drug discovery pipelines. That's fascinating. The challenge is identifying which mutated peptides might plausibly trigger immunity. What is interesting though, is how the pipeline was assembled. Normally, this type of workflow spans multiple domains genomics, bioinformatics, immunology and translational medicine and in institutional settings. Those pieces are distributed across specialized teams, document sources, and legal and technical barriers. Navigating the literature, selecting computational tools, interpreting sequencing results, and designing a candidate MRNA construct is typically a collaborative process. In this case, AI appears to have helped compress that process, pulling together data and tools from different sources. Instead of requiring multiple experts, a motivated individual was able to assemble the workflow with AI acting as a kind of guide through the technical landscape. That is fascinating. Anyway, it's a longer post, but you should go read the thing in full. Patrick Collison also chimed in. He said, according to the story, the dog's cancer has not been cured. I think it's just 50% smaller, which of course is a win. But not using the term cure is always tricky, but it does go viral. Absent all regulatory and manufacturing constraints. We could not just synthesize magic RNA, RRNA cancer cures. The technology is very promising, but it's not any kind of panacea. Yet the emergent system, emergent system of system of regulators and manufacturers is indeed far too conservative and small. Small scale experimentation is much harder than it should be. More people should read the first part of the Rise and Fall of Modern Medicine so it's interesting. Lee says ChatGPT cure cancer, make no mistakes Biomedical engineering industry yeah, don't do this. It's easy and effective, but we can't make enough money off of it. Ridiculous. It's surprising. G. Fodor says it's surprising how people are so blatantly talking past each other on this. The point is that the system of clinical trials is predicated on an assumption that a given drug will work on a cohort. What if there are lots of drugs that will only work on one person? So definitely a big desire and push for, you know, rethinking the system of clinical trials. If you're going to have personalized medicine. What does that mean? There's already a lot of people biohackers that do all sorts of stuff like this. Can't believe he wasted 2 cups of water to do this. Hashtag Banai. It is ridiculous. It's a great counterpoint to the doomers. What else is going on? Marc Andreessen chimes in I can't load the post right now. We gotta go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32 inch MacBook called the MacBook Pro Ultra Wide and it looked like this. I bought one and unlocked extreme productivity and then it wouldn't fit into my backpack so I had.

Pointingham the dog healer is coming to break down how he used AI to augment delay his dog's cancer. We're gonna be digging into. We'll first go through what actually happened. I have some opinions about this. And then we'll talk to him to get his side. Then we are pulling our delayed lightning round. We went far longer than we expected with Travis on Friday, so we are catching up on our lightning round with Tony from Sunday Robotics. And we have drew from 8 VC is a founding partner there. HEC team backed Quince at seed. It's now a $10 billion company. The Quince founder is a little under the radar. Totally. But I wanted to get this story from the 8VC team, hear how they're thinking about it and then a bunch of other folks joining. So looking forward to that. Well, let's read through Brandon Gorell's deep dive on the AI versus dog cancer. What happened so late Friday? There was a story about an Australian tech entrepr, Paul Coynningham, reducing the size of his dog Rosie's cancerous tumor by designing a custom MRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend, separating facts from the hype cycle around the story. Coining ham is an AGI pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI association of Australia. Talk about an incredible association. We don't have enough data science and AI associations globally. It's true. It's great to hear that. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Cunningham used ChatGPT to brainstorm ways he could help. And he did an interview on this and here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing, who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process. Embark, embark DNA. You could do it. Okay, well, he went to a university, probably for a good reason, probably got good data. He said, the idea is you take the healthy DNA out of her blood and then you take the DNA out of her tumor and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car and then a version of the engine at 300,000 km down the road, you can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Cunningham ran it through a whole bunch of different data pipelines. So there's a. This is something that we're going to go into. You know, throughout this story is the question of like, how much was this cure my dog cancer one shot it don't make mistakes. I don't think anyone's saying that. But very quickly there was like an incentive to amplify this into like the hype, like this crazy story. And then there was also an incentive to de. Hype this all the way. And the truth, of course, is in the middle. So that's where we're going to get today. Once the DNA sequence was produced, he ran it through a whole bunch of custom, different data pipelines to find those mutations and then use other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Cunningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to again the University of New South Wales, their RNA Institute, which used Cunningham's data, crunched down to a half page formula to create a bespoke MRNA vaccine for Rosie. Again, from the story, Cunningham ran an algorithm to inform the design of the MRNA and sent it to us. And we made a little nanoparticle and it's democratizing the whole process. They said, this is Paul Thordason, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Coynihan and his team administered the vaccine to Rosie, which was effective. One of her tumors shrank by half, though she is not completely cured. And that's just kind of the nature of cancer. Cells are dividing all the time. Everyone has some sort of low level baseline of cancer. Most dogs have a little bit. The question is, is it runaway, is it bad, is it terrible? And then it's hard to just like snap your fingers and cure it completely. But if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X the news of the story turned into a heated debate on health regulation. Yes. What is that? That was for Rosie. That's for the dog. Air horn for Rosie. The air horn for the dog. That's great. Turning to a heated debate on health regulation after biomedical engineer Patrick Heiser posted that quote, it is trivially easy, trivially, trivially easy to make a single MRNA vaccine. It's not hard. And Hank Green, a prominent YouTuber, issued something of a rebuttal, which we can go through later. A separate thread in the discourse is focused on the promise of LLMs democratizing access to medical science, with OpenAI President Greg Brockman quote tweeting the story with the caption, a small window into the opportunity of AGI. Well, Coiningham didn't literally cure Rosie's cancer with ChatGPT. As Stripe CEO Patrick Collison pointed out, it acted as a high powered search tool that ultimately helped his team get to an amazing outcome. Sort of. George Hoffman. We gotta move the goalposts. I'm ready to move them, I think. Moving the goalposts. I mean, where are we moving them to? It has to actually. You have to be able to type cure my cancer. And then from your phone it just deposits a pill that you just take. Yeah, exactly. It has to. Locally. Yeah. End to end. No, ideally. Ideally it would be not even a pill that you take. It can just create a video that you write. Pattern of light. The right pattern of light coming from. And sound. So the phone has light and sound and so the light flashes in your eyes at a certain rate. It rewires your brain and your brain decides to go kill the cancer. Yeah. And that's. We've talked about this a bunch. I think. I think it would be helpful for the industry to refocus some messaging on not AI is going to cure cancer, but human humanity is going to use AI to cure cancer and do a number of other things. Right. And so the. The bar is not just like one shotting it with a. And it sends it to a lab and you get some type of treatment in the mail. Maybe I can imagine that in the future. Right. Something to that effect. But it is an enabler, it's a tool. And this has allowed someone to become, not an expert in something, but to help somebody understand a process enough to go out and find the right experts to help them solve their problem. And I think it's incredibly inspiring. So excited to have him on the show later. So there was a chemist who works in AI and biotech by the name of Ash Jogalikar. And he had a really good summary along those lines with a riff on Freeman Dyson's 2007 New York Review of Books essay, Our Biotech Future, which we should read at some point, in which, in this article, Freeman Dyson argues that biotechnology will become small and domesticated rather than big and centralized. The full post is worth reading in full, and we might go through it, but the conclusion is particularly good. If AI continues to reduce the cognitive overhead required to navigate biological knowledge and assemble complex pipelines, the boundary between professional research and motivated individuals may begin to blur. That shift will require careful thinking about safety, governance and responsibility. But it also carries an exciting possibility. Dyson imagined a world in which biological design might eventually become something like a creative craft practiced not only by institutions, but also by curious individuals experimenting at smaller scales. Yeah, I think there's the reality of cancer treatment, from my understanding is, and this was based on a late family member that had cancer and ultimately passed away. During the process, during his treatment process, which was around a year and a half, he was getting looks at different treatments that were promising, some of which he was able to do, some he didn't qualify for just based on his personal situation, even though there was, there was a decent chance that it could have had a positive effect. Yeah. And that sort of the insane frustration that an individual feels or family feels when they're like, hey, this, you know, if, if, if something's terminal or it's looking really bad, it's progressing in the wrong direction and there's a, there's a treatment out there that isn't, that is somewhat trivial to actually make. You just don't qualify for it. Yep. That level of frustration will eventually drive more individuals, I think, to do this. Right. And so there's definitely, definitely some, like, safety. There's huge safety concerns, there's ethical concerns. These are things that we have to work through. But ultimately, I just think there's going to be so much, there's going to be enough, like human energy and just overall desire to live that people will take risks that they wouldn't take for a bunch of other more sort of like trivial sort of issues. There have been initiatives with the fda, something around. Right. To try in certain scenarios, patients rights, sort of removing some of the regulation and allowing people to make decisions like that. It does feel like the FDA's stance might need to change in this case. Like they clearly have a role to play currently and in the future where, you know, biotech becomes more democratized. But yeah, hopefully there's like some good symbiotic relationship there with the, with the broader biotech community as it get bigger. I have a similar story with someone who developed a rare illness and was able to go and read academic research at a very deep level. Didn't have a background in biotech or anything like that, but was able, this was pre. AI, was able to read like every published research paper that was at all related to this particular illness and found the world expert in this particular disease contacted the professor and the professor said, yes, you have the thing that I've been studying and I've only found five people or 10 people in my entire career that have this thing come down, I will operate on you. The operation happened, it was successful and it was fundamentally like a high agency person doing a lot of research. And if AI just acts as a search tool that democratizes that, you're going to get better results. So even if you're, even if we're not in like one shot in curing cancer, that just feels like making search easier, making research easier. Huge benefit. Yeah, Cunningham, the guy in Australia could have done a lot of this 10 years ago. He just would have needed to spend, I'm sure, a bunch of time and yeah, that's the thing. Libraries, everything you do all these things you can do manually. You can just get a guy for that. Yeah, you can get a guy or I mean, you don't even need a spreadsheet. You can calculate the math by hand, but these things speed things up. So it's been a good time. Let's read through Ash's post. But first let me tell you.