LIVE CLIPS
EpisodeĀ 3-31-2026
Since you started meeting with Demis, because in some ways, I have to imagine you kind of. Maybe it hasn't been that surprising at all, even though a lot of the growth is impressive. But do you feel like you had a view into the future from those first conversations in the pub? Yeah, I mean, I think I was lucky that the meetings were bookended by going to see Dem is sort of maybe the third meeting or something. ChatGPT by this point had gone viral, and him saying to me, okay, this is war. You could see his competitive side come out, right, this is war. These guys, he said, have parked the tanks in my front yard. I'm fighting back. So you could see that. And then after that comes the merger between Google Brain in Mountain view and the DeepMind team in London. So the two kind of halves.
At the Apple headquarters or something. It's just a different vibe. Could you ever see Demis as CEO of Google or would he have to do too much paperwork? This gets to the heart of the dichotomy about Demis. It's so difficult because he's so many different things at once. He is this Nobel prize winning scientist who would love to just do pure research and he often would fantasize to me about hey I want to retire to the Princeton Advanced Institute of Advanced Studies and do what Einstein and Oppenheimer did before me and go there and I think he really means it when he says that at the same time he also wants to be in command of an AI lab and he likes the capitalist competition so if he had the opportunity to be chief executive of the whole of Google I suspect he's too competitive to say no but who knows I mean there is that science side so it's genuinely unpredictable. He probably doesn't know himself Were you left personally.
Of repurpose into this. What's your theory? Right now it's obviously day, day one, but do you think in the future we'll get, you know, more volatility? Because you have like financial institutions that are effectively using like agents or algorithms to do trading and then you also have retail. So when you get like a new CPI print, you know, you have, you know, even, even, you know, additional trading activity off of these single events. Like do you think this is something in the future that everyone will effectively have like a handful of agents running just naturally in the product and then some people be like maybe prosumers, people that are more into it will have many or at some point. At what point is it old fashioned to be just buying a stock yourself, even with a button? I actually think we will look back at tables and buy buttons and feel that's a little antique, maybe already 12 months from now. But I think the effect is things will get priced in faster for sure. On the institutional side they've, you know, they've used some version of AI for the longest time. Right. But then at the same time retail have gone from being like 5 to 25% of the market. And on the retail side folks haven't been as fast to react always. Right. They haven't been that disciplined. They're not necessarily glued to the screen 24:7 and so, you know, they can't always react as quickly as they want to and agents obviously change that. Right. And so I do think that there might be, I mean it's a little bit like whether it's crypto prediction markets, there's always a little bit more of sort of an alpha opportunity or an ARP opportunity in the early, early days. And then over time it becomes more mainstream and that kind of fades away. And I would suspect that this follows something like a similar pattern at least. Yeah, I've just been thinking.
This process about just our AI future, broadly net positive impact from AI. I mean, clearly there's a lot of upside, especially in medicine and pharmaceutical discovery. I think you have to be honest and say there's also downside, significant downside. And as I continued to do the research for this project, I became more worried about the downside because, you know, people like Geoffrey Hinton, I went up and spent two hours in his kitchen in Toronto and sat there debating not necessarily whether machines would be more intelligent than humans, obviously, that they will be, but whether they. Whether machines have a motivation to harm humans. And he's very persuasive in arguing that they will. I mean, my point was, look, humans evolved over centuries to survive, to pass on their DNA. We're hardwired to survive. That's why we fight each other. Machines aren't like that, so why would they attack us? Why are you so worried, Jeff? And Jeff is like, well, imagine you've got this super powerful AI and it's going to be attacked by the enemy AI and you have to defend it. So you tell the AI if you see a cyber attack coming, you've got to fight back, you've got to defend yourself. And now all of a sudden, you've given your AI a survival instinct. And so don't tell me that evolution has to happen as it happened to humans. The evolution can happen in a machine way, but these systems are going to want to survive and they're going to be more intelligent than us. So we're in trouble. And I think you can't dismiss that. So I'm kind of both worried and excited at the same time. And I think that's how humans generally respond to all technology. And if we didn't take that trade and move forward with both the excitement and the scariness of technology, if we didn't take that, we'd be still living in caves. Yeah. So in some sense, the story of Demis is like the story of all of us, but magnified 10x. There's a document.
Incredible for him to stay in the research mode, build the research organization. And yet because of scaling laws, we're in this weird regime in this paradigm where sometimes doing a deal to Marshal an extra 10 billion of compute actually does unlock a new capability and is almost in the research track. And I'm wondering how you perceive the relative importance or Demis perception of the relative importance of these sort of like business dealings that might be more critical path to AGI than the demis of 2021 might think. You know, I would say that the single most important business relationship in the world today is between Sundar Pichai, CEO of Google, and Demis Asabis, who's running the AI brain trust. Because Sundar has Demis back. He deals with providing the resources, supporting the notion that you're going to spend all these tens and maybe hundreds of billions on compute. That's Sundar's, that's what he delivers. And he gives Demis the oxygen to then just go do the science. And sure, he has to build products, but I think he said to me several times, temis, we've got to a point where building a product like Gemini is in fact advancing the progress towards AGI. There's no tension between the two. If you had stopped your AGI research 10 years ago and taken a sidetrack to build some widget, yeah, that would have been a waste of time. But now that it's so mature and you're actually to build the next LLM, you've got to kind of figure out a reasoning model. Then it's going to be agentic and then you're going to be scaling it even more. And all this stuff that we've seen, this is genuine scientific progress as well as product advance. Yeah.
I'm fighting back. So you could see that. And then after that comes the merger between Google Brain in Mountain view and the DeepMind team in London. So the two kind of halves of Google's AI talent base are united. And then I think you get what a business school professor is in the future going to write about. It's kind of like a textbook case in how you make a merger successful because everyone knows that mergers are hard. And when you do it with eight time zones between the two teams, one in London, one in California, and you're doing in the middle of this knockdown, drag out, capitalist fight over building LLMs, this is going to be. Most people said this is going to fail. And I would come to Silicon Valley while I was doing the book, check in with my friends at different venture capital shops and they'd all say, game over. OpenAI has won. The surprise is that within two and a half years, Demis Hasabis's Google DeepMind model Gemini was doing better on the rankings than the OpenAI models. So that was an incredibly fast comeback. Yeah. How do you think about.
Do you feel like you had a view into the future from those first conversations in the pub? Yeah. I mean, I think I was lucky that the meetings were bookended by going to see Dem, sort of maybe the third meeting or something. Chatgpt by this point had gone viral. And him saying to me, okay, this is war. You could see his competitive side come out, right, this is war. These guys, he said, have parked the tanks in my front yard. I'm fighting back. So you could.
Business Walk us through the launch today. Yeah. So today we launched AI Agents for Investing. Yes. The easiest, safest way to put AI agents to work directly inside your portfolio. And the way it works is pretty simple. There's now an agents tab directly within the public app. You chat with an AI to set up agents that monitor markets, move money around and even execute trades for you. All within the app. Right. So there's nothing to install from a security standpoint. Obviously everything stays in a safe, controlled environment because it's within the brokerage all Axios. No, do not do that. That is such a niche joke. Everyone's going to think you're talking about the publication. Yeah, no, but I mean, it's been really fun. I have a bunch of agents running now on my account. It's been really awesome to see how it's changed my behavior as an investor. Right. So, John, relax. I like the sound of that. I knew you're going to do that. We got little John in the studio. No, so, so, so make makes total sense. What, like if people are like already signed up, which they should be. How, what should they. What's the first thing that they should try to like, set up or experiment with? Like, what was your first few agents that you've set up and continue to keep running? Because I'm assuming you can effectively like run them, you can retire them at different points. Exactly. The first one I set up was one that checks the oil prices before the market open and buys protective put options every day. As a hedge. Sure. If there's a spike in those, I call it the I'm tired of seeing red due to war. And so today that didn't fire. Thank God. You know, I have another one that just looks at my bank accounts and automatically sweeps any cash in excess of a certain amount into my bond portfolio. So I'm always yield maxing. You want to be yield maxing? Always. Especially while rates are still high. Yeah, that makes sense. And, and then I got some more advanced stuff. Like I got one that scans the markets for opportunities to write covered calls. Okay. Across my top position. So if there's a low risk opportunity to make like 20 grand a month. Yeah. Selling options, premiums, I instructed it to just go ahead and place those orders. So that just rolls every time, all the time. And I don't have to think about that exact strategy was the first thing a private wealth manager ever pitched me in my career like a decade ago. They're like, I. And they had like a guy that did it and you had to have a lot of money to, like, access that. And there were minimums and stuff. You should probably check on that guy after today's launch. No, no. So there are obviously incredibly advanced things that you can.
Build intelligent real time conversational agents. Reimagine human technology interaction with 11 labs. So moving on, more hacks last night. More leaks. What's going on? Last night POD code, source code was leaked via map file in the NPM registry. There's just a link to. Wait, someone's just actually. Do not click a link. If somebody ever says, hey, I got some really great source code here, just click this link. Probably don't click it. Let other people screenshot it. There's plenty of meta analysis over here. Seems. Seems messy. Seems unfortunate. Heart goes out to the folks who are dealing with the situation at the same time. Codex is open source. It's not the end of the world. But it did reveal a bunch of things about the roadmap and also some of the internal April Fools. That is the worst part. We love a secret surprise April Fools joke. I love a good joke. And nothing spoils a joke like hearing about it a day early. But much more importantly, there are lots of other critiques of the way CLAUDE code is implemented. What are the I don't think this hurts their business at all because people are using CLAUDE code to make other products and then also having to take basically a fork of CLAUDE.
Conversational agents reimagine human technology interaction with 11 labs. So moving on, more hacks last night, more leaks. What's going on? Last night POD code source code was leaked via map file in the NPM registry. There's just a link to. Wait someone just actually do not click a link. If somebody ever says hey I got some really great source code here, just click this link. Probably don't click it. Let other people screenshot it. There's plenty of meta analysis over here. Seems messy. Seems unfortunate. Fortunate heart goes out to the folks who are are dealing with the situation at the same time. Codex is open source. It's not the end of the world. But it did reveal a bunch of things about the roadmap and also some of the April Fools. That is the worst part. We love a secret surprise April Fool's joke. I love a good joke and nothing spoils a joke like hearing about it.
So moving on. More hacks last night. More leaks. What's going on? Last night source code was leaked via map file in the NPM registry. There's just a link to. Wait, someone just actually do not click a link. If somebody ever says, hey, I got some really great source code here, just click this link. Probably don't click it. Let other people screenshot it. There's plenty of meta analysis over here. Seems messy. Seems unfortunate. Heart goes out to the folks who are dealing dealing with the situation at the same time. Codex is open source. It's not the end of the world. But it did reveal a bunch of things about the roadmap and also some of the eternal April Fools. That is the worst part. We love a secret surprise April Fool's joke. I love a good joke. And nothing spoils a joke like hearing about it.
No, this is my mark. I will believe that it's real if I see an astronaut. Put three fingers in front of their face. Yep. Because this is the one thing that the AI can't do right now. If you're ever on a zoom call with someone who you suspect of being fake, a scammer who said, hey, let's get on zoom, let's talk about some financial investment opportunity, and you. And it looks like someone you think is the person, but you suspect that it might not be. And they will be able to show you. Look. Look at the fingers. The fingers are perfect. It's fine. It's fine. That's because this part is not AI Just the face is AI. This is the deep fake stuff that's happening. So what you have to do is you have to ask them to hold up three fingers. They'll be like, yeah, three fingers. This is fine. Right? I satisfy the task. You got to say, no, put the three fingers in front of your face. Because if you put the three fingers in front of your face, the AI gets confused and it breaks the deep fake that's happening underneath. So you gotta go like this. Show some depth of field. You have three fingers in front of the face. This is the trick. This is the only way you'll survive in the future. Be careful.
Where we can take it further and then obviously and Dylan's from Long Island, New York, so John is going to be I'm going to do it in a Dylan Abrascado impression. Memes are changing. That became abundantly clear during the Oscars a few weeks ago when Conan tried to create a New Leonard Leonardo DiCaprio meme that's just like UFC announcer to go alongside the classic Leo memes. In doing so, especially by using tfw, that Feeling when and the blocky white font that defined early Internet memes, he inadvertently demonstrated that the meme templates millennials grew up with have become increasingly stale, even cringe. That's a good point. Instead, AI generated videos are the new meme template that every network and studio should be focusing their launches on. Look at what happened. Look at what's happening with the Harry Potter reboot. When the trailer first dropped, the reaction to the new Snape played by Ghanaian. Sorry, Ghanaian. He's from Ghana. Ghanaian English actor Papa Isidu was predictably and unfortunately negative. According to the LA Times, he received death threats since being cast in the new role. But after a few incredibly viral and well produced AI videos, one an original Snape vs Black Snape MMA match and another AI generated rap video and another dripwarts the school of Drip, the narrative has started to shift. Have you seen any of these? I think I've seen Drip warts. But can we pull up the original the quote Unquote original Snape vs black Snape MMA match? Because I have not seen this one and I think it is illustrative of what Dylan is talking about here. Snape v Snape in the UFC ring. While we pull that up, let me tell you about Plaid Plaid Powers. The apps used to spend, say borrow and invest securely connecting bank accounts to move money, fight fraud and improve lending. Now with AI. And let me also tell you about Restream 1 livestream 30 plus destinations. If you want to multi stream go to restream.com so let's take a look at Snape v Snape in the UFC ring. Any luck? Here we go. I just dropped it in. Cool. And we have a few others here. The videos have amassed tens of millions of views and on Dylan's timeline at least sentiment around both the character and the reboot has done a complete 180. Here we go. Really photorealistic. Does this Are there any red flags here? As a UFC enjoyer, does this feel like a proper UFC actual video quality? So bulky. The video quality is insane. Wait, but old Snape Won in the fight. Yeah. Okay. But I think it just. I. Okay, wait, how. How. How do you know that? I think it just sort of like, makes the characters more entertaining, more fun. Shows you that this is just creativity. At the end of the day, this is just like, you should not be so up in arms about something that's a movie. Like, it's entertainment. And here's some more entertainment. And so you're. You're adding entertainment to the discussion, and people are enjoying that. There's another AI generated rap video about the new Snape, which we can pull up a little bit of here. AI meme videos are inherently viral and driving real awareness in a way traditional memes no longer can. Not just because they're novel and more entertaining, but because a single AI clip can travel further and compound harder than traditional meme formats and social feeds that now heavily favor video. This suggests. Yeah, it's interesting on. On X, it's still very easy for an image to go viral, but if you think about, you know, Instagram, YouTube, like a standalone image, just can no longer actually get that escape velocity. I mean, what about Dripped Out Pope? Remember that? Yeah, a little bit. But people are just spending so much time in the short form feeds, and those can go in there, but there's certainly. Yeah, yeah. I mean, I guess even some of the Dripped Out Pope or. What was it? Was it Balenciaga Pope? I don't remember what the name was, but Dylan says this suggests a new playbook for marketers, especially in entertainment. If you're about to drop a trailer for a new movie or show, you need to be thinking about your rage bait character, the one people will latch onto, Remix with AI and build around. Conan tried to force a Leo meme down our throats at the Oscars. Didn't see that because I was sleeping, but this might have worked 12 years ago. That playbook is over. Today, enraged fans and communities will, if you're successful, take your characters or moments and turn them into something much bigger. Entire cinematic universes. Yeah, yeah. I'm just very impressed by the overall quality of those outputs. The Oscar selfie, I remember this. This, I think, became the most liked image on Twitter at the time in 2014. Briefly, that image. This is the canonical clout bomb. If you're a fan of Bradley Cooper, you like it. If you're a fan of Meryl Streep, you like it. You're a fan of Brad Pitt, you like it. And so you're amplifying all of the ultimate collab post. And this has become a format that's been used time and time again. It's effective. We did a little bit of it at the Super Bowl. It's fun. It works. But now the future is AI let's pull up the dripwart school of Drip video. I want to watch this one because I saw a clip of this. I didn't see the whole thing. Let's see if we can play this. That's Harry Potter. Are you really Harry Potter? My gift. Type, type, type, type. None of that. None of that, Broski. We're all here on the Maybach Express for one reason and one reason only, and that's to go to Drip. Watch the school of Drip. The Maybach pulling the train is pretty good. And there's. And there's the new Snape character. So, yes, very effective. I was reflecting on this and thinking about how it's not just AI videos that are unlocked as the new meme format. Like, 20 years ago, video editing was extremely difficult. Like, you had to do it on a desktop. You had to have a piece of software that probably cost a lot of money. It was not.
Intelligence out into the real world. And each of these verticals are facing really, really different problems. You take ag, you know, with a tracker behind me, the average American farmer is 58 years old. There's nobody coming to replace that person. And so what is going to happen? Because, you know, if you take that person, their kids have left and they're often not coming and taking over the farm like maybe in previous generations. So that farmer needs, you know, we don't need to teach them how to use Claude code. That's not what's going to change the farmer's trajectory. What's going to change the farmer's trajectory is the machines are intelligent and they're working harder and smarter on their behalf. And so he can run an entire farm with a, you know, with a swarm of machines. And that's not, you know, that's not too far into sci fi. One of the key components here that, you know, we're doing and we believe is you need to abstract that hardware and software away. We as technologists, you look at, like, your laptop and your phone, and you kind of take for granted the miracle that exists. Android runs on thousands of hardware devices flawlessly. Yeah. So that's also something that applied, does. We're just abstracting the hardware and software. Once you do that, you can make every machine, you know, intelligent. Have you tried to estimate the economic impact? Assuming, you guys, you know, stay at the, you know, at the current kind of.
Glitch. That was. That was vfx. That was AI. No, this is my mark. I will believe that it's real if I see an astronaut. Put three fingers in front of their face. Yep. Because this is the one thing that the AI can't do right now. If you're ever on a zoom call with someone who you suspect of being fake, a scammer who said, hey, let's get on zoom, let's talk about some financial investment opportunity, and it looks like someone you think is the person, but you suspect that it might not be, and they will be able to show you. Look. Look at the fingers. The fingers are perfect. It's fine. It's fine. That's because this part is not AI. Just the face is AI. This is the deep fake stuff that's happening. So what you have to do is you have to ask them to hold up three fingers. They'll be like, yeah, three fingers. This is fine. Right? I satisfy the task. You gotta say, no, Put the three fingers in front of your face. Because if you put the three fingers in front of your face, the AI gets confused and it breaks the deep fake that's happening underne. So you got to go like this. Show some.
Machine, you know, intelligent. Have you tried to estimate the economic impact, assuming you guys, you know, stay at the, you know, at the current kind of improvement rate or accelerate as the technology kind of starts to diffuse. And in some of these industries like trucking and mining and agriculture, like, what, what are the downstream impacts? I mean, there's such a debate right now around what, what, what impact will AI have on the economy? And so much the economy is like moving physical things around, producing things. Yeah. So let's, let's, exactly, let's separate a little because economy is such a generalization. So yeah, when you're talking about like, you know, code complete and white collar work is very different than, you know, trucking, where there is a huge labor shortage. It's very different than in mining where, you know, people don't want to go live in kind of remote areas doing 12 hour shifts. I mean, literally, labor shortages are preventing construction companies from, you know, collecting billions and billions in revenue. So these are industries where I can't get there fast enough. Yeah, it's a very different calculus than a kind of, you know, I think what the normal narrative is. And, and then we're super obviously excited about that. Let's take defense as a particular example. It's a very salient example. We don't need more warfighters in harm's way. We need less war fighters in harm's way. And no war fighter wants to go out into that ecosystem where autonomous autonomy is really becoming the dominant thing. And so, so I think the way to think about this impact in the physical world is it's a lot less resistance. There's a lot more pull. Now the first question you asked is the size of impact. I don't want to sound like I'm pitching my own book here with the masking. I'm asking you. I want, I want the biggest number. I want the biggest number. The numbers are absurd and ridiculous. But I can tell you this much. If you think about, you know, the way I think about, you know, I used to be a Y combinator before I was a COO and you know, ran the firm and funded lots of interesting companies. And one of the analogies I used to use to help founders understand market potential, market sizes. Yeah, I grew up in Detroit. You're sitting in the Detroit Metro airport and you're sitting in a gate and you look around, how many of those people are like really deeply using Claude code? I mean, frankly speaking, not many. Not will probably be using something like ChatGPT, some variant of that, maybe Gemini, but How many of those people drive? How many people work at construction sites? How many of those people ride in buses? How many of those people serve in our armed forces? The point is a much, much larger group. And I feel a little. Again, the engineer in me feels a little awkward saying these kind of, you know, pitching these things. But I think the market for physical AI is way, way bigger purely because of the surface area is much bigger. And it's compounded by the FA of the way that the way technology diffuses with phones and laptops creates this, like, rabid, you know, competition that you see in, you know, you're seeing in all these kind of subspaces, Right? Yeah. In physical AI, you got to kind of know what's going on in the car business. And I'm not saying, you know, I'm not, you know, gatekeeping and saying, yeah, you got to go to the General Motors Institute to build technology for the car business, but you bet your bottom dollar it helps. And. And we're doing that across a bunch of industry. I think it's. I think it's. You know, I'm. I'm as confident about the company as ever before. You know, the question. We always get asked this question, why the hell did you raise all the.
Really, really different problems. You take ag, you know, with a tracker behind me, the average American farmer is 58 years old. There's nobody coming to replace that person. And so what is going to happen? Because, you know, if you take that person, their kids have left and they're often not coming and taking over the farm like maybe in previous generations. So that farmer needs, you know, we don't need to teach them how to use Claude code. That's not what's going to change the farmer's trajectory. What's going to change the farmer's trajectory is the machines are intelligent and they're working harder and smarter on the, on their behalf. And so he can run an entire farm with a, you know, with a swarm of machines. And that's not, you know, that's not too far into sci fi. One of the key components here that, you know, we're doing, and we believe is you need to abstract that hardware and software away. We, we as technologists, you look at like your laptop and your phone and you kind of take for granted the miracle that exists. Android runs on thousands of hardware devices flawlessly. So that's also something that applied, does. We're just abstracting the hardware and software. Once you do that, you can make every machine intelligent.
Trade offs that come with them. And by the way, even if you choose to implement one, you need to test it and you make sure it's secure. All this stuff. So have you tried to, have you tried to quantify or guess estimate what the quantum discount rate is on bitcoin right now? Because it's an interesting thing where like if you own a lot of bitcoin, there's a bunch of people on the timeline today talking about this. They don't have an incentive to like really freak out and spread the narrative, but they have some incentive to say like, hey, we need to have a conversation, we need to make progress on this. But do you think that's factoring into price at all? Totally. The way I would put it is I think if this risk didn't exist, bitcoin would be priced significantly higher. So I think exactly what you said is right. You're like, oh, I'm not going to sell my bitcoin because it's maybe not right around the corner and I'm hoping people fix it. But I also think there's people that would maybe enter and they're like, Chamath has said this exact thing. He's like, hey, is this really digital gold? Can we without with this quantum threat hanging over everyone's head? So I think if that threat was removed, then you remove this cloud over that ecosystem and potentially you'd have a lot more people coming in and therefore price would be up. Yeah, it's easy to imagine as you approach that 2,029 mark, more selling pressure.
We've done that too. We collaborate with the DF and we're doing, we're getting ready to launch our own post quantum wallet as well. Yeah, talk about the information flow. How much of the work that you do, the work that will be done by the Bitcoin Foundation, Ethereum foundation, all the different developers, how much of that is open source by default or licensable or just can be understood by other parties and implemented very quickly? Like, how should we expect diffusion when once this problem is solved, to actually roll out? Will it just be like, oh yeah, like we're just following the Solana standard and so we're just going to mirror that over onto whatever chain we're working on? Yeah, I think there will be diffusion, I think there will be sort of consensus, if you will, around a certain subset of post quantum algorithms. But I don't think it's just one and done because Solana is a very different system than Bitcoin. Right. Bitcoin's digital gold. It's sort of meant to be slow, there's no apps on it. Solana is meant to be fast. Right. And so the cryptography that works for Bitcoin might not be this cryptography that works for Solana. And this is actually was kind of the results of some of the experiments we ran with Solana. And look, this is one of the challenges, right? And this is again why we keep saying this is like time to start is now, because we don't know how long it's going to take to migrate. Because these new algorithms, there's trade offs that come with them. And by the way, even if you choose to implement one, you need to test it and you make sure it's secure. All this stuff. So have you TR.
Do think, especially in light of these two papers, these two breakthroughs, you just can't stop or you just can't wait anymore before starting that process. What about the rest of the digital world? Because if, if, if, if Bitcoin is having problems, then so many other kind of core institutions and companies, organizations, I imagine, would have issues as well. Maybe, maybe because they are centralized, there's easier to react, easier to kind of lock things down, but still need to upgrade overall encryption. Yeah, there's no doubt that other institutions need to upgrade, but in my mind, there's also no doubt that blockchains and digital assets are just the most vulnerable. I mean, one reason is obvious. I mean, Satoshi is bitcoin. So if the founder of bitcoin, who we think has gone away or died or something, they have a bunch of their early Bitcoin that hasn't moved, there's a bunch of lost coins. You know, all in all, you know, it's about maybe 15% of all of bitcoin supply is estimated to be lost. I mean, that's hundreds of billions of dollars potentially in market terms. So that's just a huge incentive that, like, let's take, let's take the counterexample of someone wanted to hack into a bank or something. As you pointed out, banks are centralized. They can kind of react. Also, the cryptography, the way that banks implement this cryptography is just kind of one of many layers of security. Right. So it's kind of. This breaks like theoretically, someone tried to wire all the money out of my account, my bank call. Yeah, no, they literally have tape drives where, you know, they have, they have cold storage. They print things out and they have a ledger and they can potentially roll back, which is crazy to think about. But, like, they could. If there was like a catastrophic hack, they could be like, look, everyone's just going back to yesterday's accounts and, you know, that's better than the chaos that we're in. That's it. And that's not true for bitcoin. All I need is one signature and all of satoshis or coinbases or finances. Bitcoin is mine. There's no fallback. There's no anything. And that's how it was designed. Right? That's the point. That was the point. Permissionless finance, that's the challenge. So what is the.
Pursue this. Solving this problem very vigorously is because everything's on the line here and we have to solve it for these chains like bitcoin to have a future. Yeah. Is there generally low optimism right now that bitcoin developers will be able to react quickly enough? What's the statement about, like I think Churchill said about democracies or the Americans may was like they'll do the right thing when every option is exhausted. I think this is. Look, I think this is true of decentralized networks like bitcoin. I mean their greatest strength is the fact that there's no single party that says how it works or how it should work. Right. And this was encoded into how it was built by Satoshi as a reaction to the great financial crisis. Right. So that's a great philosophical strength in the face of a crisis like this that demands a massive technical effort to overhaul. It's a daunting challenge because unlike say Google, which Google has said they're going to upgrade all their systems by 2029, that's just someone at Google can make that decision. Snap, right. In Bitcoin, because it's distributed community, everyone's kind of first has to agree there's even a problem, then everyone has to agree on the solution. But I think there's examples of places where blockchains like I'll take Ethereum have done amazing things, right? So one is they transition from an old system of consensus called proof of work to a new system called proof of stake. It took four years to be sure, but it involved thousands of people all over the world. And they did it. They did it. The blockchain's been running. Ethereum is second largest blockchain by market cap. So I don't think it's impossible, but I do think, especially in light of these two papers, these two breakthroughs, you just can't stop or you just can't wait anymore before starting that process. What about the rest of the digital world? Because if bitcoin is.
No one can put a timeline on it, but how fast do you think you can close that gap is the question. Okay, is it possible Jordi was throwing out the idea of someone having a secret quantum computer going around the blockchain, siphoning bitcoin from cold wallets that haven't wasn't implying that it exists yet, but implying the incentive of if somebody were to create one of these. But the way you're reacting, I'm imagining it's like if somebody does it, it'll be Google first, which is maybe a good thing. I don't know. It's like, it's really hard to know how it's going to play out. I wrote a whole blog post on our. On our blog on project11.com, people can check out called Quantum War Games. And it was really fun because it's exactly. It's like the what if scenarios. Right. You know, because why do people want quantum computers generally? Well, they're great for science 2. Like, you can imagine, governments that want to do espionage might want the ability to break cryptography, too. They probably don't want to reveal what they have. Certainly not if it's China or Russia. Right. And, you know, but in private companies, maybe not Google, maybe some of these pure play quantum companies, like, how are they going to make money? Well, one way would be to recover satoshi's bitcoin as if it were buried treasure. Right? Like, oh, buried treasure. Satoshi is not here. It's mine now. Right. So, I mean, that could be another scenario. So, look, I think that there is just a whole bunch of uncertainty about how this is going to play out, about who's going to execute the attack, about how long a quantum computer will take. And again, because blockchains like bitcoin fundamentally rely on this cryptography, it's existential for them. That's one of the reasons I founded Project 11 and we pursue this. Solving this problem very vigorously is because everything's on the line here, and we have to solve it for these chains like bitcoin to have a future.
We cannot wait to migrate anymore, that people are going to start paying attention. Okay, what actually changed? Because it doesn't seem like the, like the number of logical qubits or physical qubits, like, that seems to be growing exponentially. But even when you trace out the curve, you're five, 10 years away from what we thought we needed. So is this a new algorithm, a new stack of code, or is it new math? Like, what changed that? We got this 20x increase in efficiency in terms of cryptography breaking via quantum computers. Yeah, a couple of things. So first off, these two papers are not necessarily about a quantum computer that's bigger or more capable. Right. So this is. They're about what is what it takes to break cryptography. Right. And so what changed? So one of the things that changed was that, interestingly, physicists and, you know, kind of quantum cryptographers that looked at this problem for a long time studied an algorithm called rsa. It's not worth defining, but it's kind of an older cryptographic algorithm. But that's not what really any blockchains use. Right. Because, you know, RSA keys are very large. Right. So it turns out this was kind of the key, one of the key upshots of the Google paper. It turns out that if you actually focus on the cryptography used by Bitcoin, Ethereum and other networks, it's actually way easier to break than they thought it was compared to rsa. And so that is one of the major things. The other big breakthrough, and this is from the other paper from Caltech, is that quantum computers, as you guys may or may not be aware, as your audience may not be aware, are kind of. They're very fragile generally. So to be useful, they need to have what's called error correction applied. And that error correction can kind of result in a lot of overhead. You need to have tons of physical qubits to get to. You mentioned logical qubit to get one logical qubit. Well, this Caltech paper basically showed, hey, we have some new ideas to do error correction. And it turns out if we apply those, we don't need hundreds or thousands of physical qubits. Maybe we just need a handful to make one logical qubit. So that the title of their paper, actually the headline is, you may only need 10,000 physical qubits to break short to run Shor's algorithm. And by the way, they demonstrated last year 6,000 qubits. Okay. So, yeah, that's, you know, no one can put a timeline on it. But how. How fast do you think can close that gap. Yes, is the question. Okay. Is it possible Jordi was throwing out the idea of someone having a secret quantum computer going around the blockchain, siphoning.
So just over I guess almost a year and a half ago and what was what was the general dialogue around quantum and the risk to crypto at that time that it wasn't real, that quantum computers were always going to be 20 years away that you know, that no one had to pay attention there were bigger things to worry about. Honestly I feel like that's slowly changed and I think today's not just so there's a paper from Google, there's another paper out of Caltech both dropped on the same day both effectively lowered the bar massively that a quantum computer had to clear to be considered cryptography cryptographically relevant to threaten Bitcoin. Right. So that was the breakthrough and I think this is a watershed moment where really at this point when Google, the head of the Ethereum foundation and a Stanford cryptography professor all pound the table and say we cannot wait to migrate any more that people are going to start paying attention. Okay, what actually changed? Because it doesn't seem like the.
Think if that threat was removed, then you remove this cloud over that ecosystem, and potentially you have a lot more people coming in and therefore price would be up. Yeah. And it's easy to imagine as you approach that 20, 29 mark, more selling pressure, more concerns if meaningful progress isn't made in the next two years. What countries have the most kind of advanced quantum projects outside of the U.S. i can guess, you know, China's, you know, investing heavily here. Do they have their own retail quantum companies that are trading like crazy? What's going on? Yeah, first off, I think it definitively the leaders, both companies and research, is American. So I think we should be proud to be an American here. But look, I think one interesting thing about the way China has chosen to attack this is they've made quantum computing a priority. And what that means in China is, you know, it's like there used to be a lab at Tencent and Baidu and a few other places. And at some point, a Chinese Communist Party official came in and said, guess what? You guys all work for us now. And guess what? You're all working together now. And guess what? You're not allowed to talk about it anymore. And that's the state of things. It's kind of a. I don't want to say Manhattan Project, but it's like that level of secrecy in China, and there's a legitimate question around how far back they are. So the best estimates that we have from quantum, like, we have a quantum physicist who's an Advisor to Project 11 that, you know, tracks generally resource estimates across the world. And their view is that China may be six to 12 months behind at most. And so this is. Yeah, exactly. That's not that far. Yeah, that's really, you know, and so can we expect, you know, quantum computer in the hands of the Chinese Communist Party that maybe is more willing to crush dissent in places like Hong Kong, to care as much about the philosophical principles of Bitcoin and decentralization if it serves their purposes to do otherwise, I think we can. And so I think, again, back to the fundamental problem, uncertainty. Right. And it's better to be safe than sorry. So we need to basically prepare today to prevent the crisis tomorrow to keep the trust in these systems. Well, thank you for.
Of Devin, the AI software engineer. Crush your backlog with your personal AI engineering team and let me tell you about Label Box. Oops, sorry, Label Box, RL Environments, Voice Robotics, evals and expert human data. Label Box is the data factory behind the world's leading AI teams. So there is a whole bunch of hack news going on. We're in a very weird week in terms of the news cycle because it's spring break and so a lot of executives at big tech companies are like, don't launch while my kids are out of school and we're going on vacation. I actually think this is my real theory. So we're in a little bit of a slow news week and you can see that like the Journal is covering announcements that happened last week. They're talking about Sora, they're talking about Disney, they're talking about things that are more like reflective. In Strathory, Ben Thompson has sort of a 50 year retrospective on Apple. It's not driven by a news item. Like it's not like Apple launched a new product this week. So Ben Thompson is taking a step back and reflecting. It's a great piece, but it's not exactly news driven because there isn't that much news coming from big tech companies, coming from the labs, et cetera. But there are a ton of crazy hacks starting with Axios. There is an active supply chain attack on axios. One of NPM's most depended on packages. So if you have been vibe coding, Axios is a a package that helps with HTTP requests. So it gets sucked into all sorts of different projects and if you upgrade it to the latest version, you basically got a virus with that. And if that's running in the cloud it's building and that's probably maybe bad because it could steal API keys or SSH keys. It could do a lot of things could wreak havoc on your system. Also if you built this piece of software and you included the contaminated Axios installer or package locally, it could potentially weasel its way out of your local environment and get onto your desktop. It's a virus, so be careful out there and I'm sure people will be responding. The recommendation from Feros who sort of broke the news over at Socket Security, is that if you use Axios, pin your version immediately and audit your lock files. Do not upgrade. Socket analysis confirmed that this was malware. Plain crypto JS is an obfuscated dropper loader that deobuscates embedded payloads and operational strings at runtime, dynamically loads FS OS and Exec sync to evade static analysis. Executes decoded shell commands, stages and copies payload files into OSTEMP and Windows program data directories. Deletes and renames artifacts post execution to destroy forensic evidence. So very risky, I would say. Like, if you have installed this, you should just, like, freak out, basically. And if you break your computer, that's like the first thing you should do. Just, like, try to slam it. Yeah. Take the computer, throw it in the lake, throw it in the ocean. That's how you should start. I concur. I mean, practical. Yeah. I mean, there is going to be some sort of, like, power law response here where of the people that are victims of the attack, they will go after the most vulnerable with the highest, like, ransomware potential. And I think we're seeing that with one company. I believe Mercour was targeted, but I don't know if that's. Because I don't. Was that. Yeah. My understanding is that. Yeah. The crazy thing is you have. You have this, like, Claude code leak that. That was completely separate, even though I do believe they use Axios in Claude code. Okay. Saw something on that. Sure, sure, sure. And you have the Mercour leak, which is. Well, it's not a ransom ransomware. Someone stole some data. Yeah, they stole a bunch of data and now they're trying to, you know, get bids on it. We'll get to that in a little bit. Okay. And then there's. There's this Axios supply chain attack. Yeah. Anish had a little bit more context. He said a tiny piece of code called Axios runs inside almost every app on your phone and every website you visit. Developers download it 100 million times a week. A few hours ago, someone poisoned it with malware that hands an attacker full control of your computer. If you've never heard of Axios, that's normal. It does one boring but important job. It lets apps talk to the Internet. When a website pulls up your feed or an online checkout processes your card, Axios is probably doing the work underneath. Over 173,000 other code packages plug into it. It's everywhere. The attacker stole a lead developer's login for npm. Think of it as an app store, but for code that programmers use. Once inside, they swapped the developer's email to an autonomous ProtonMail account and uploaded the poisoned version by hand. That jumped past every security check. The project normally runs before new code goes live. And this was not a rush job. The stackers staged the malware at least 18 hours before pulling the trigger. They built separate versions for Windows, Mac and Linux. They poisoned both the current version and an older 1 Within 39 minutes of each other, casting the widest net possible. Once the malware ran on a machine, it deleted itself to cover its tracks. The trick was smart. They never touched a single line of code inside Axios itself. Instead they tucked in a fake add on called Plan Plain Crypto js built to pass as a well known trusted library. It copied the real library's description and author info so nothing looked off at a glance. When a developer installed Axios, this fake package quietly ran the malware on its own. When a smaller package called UA parser JS got hijacked back in 2021 with about 8 million weekly downloads, the security world treated it like a four alarm fire. Axios has 100 million over 12xe exposure with 173,000 packages depending on it socket. The security firm that flagged this caught it in about six minutes. That's fast, but six minutes is still plenty of time for automated systems at companies everywhere to pull and install the bad version before anyone can react. If you or your team run Axios, freak TF out no lock your version to 1.14.0, change every password, API key and access token on any machine that installed the compromised update, and check your network logs for connections to sfrclak.com or the IP address. 1421-120-6673 Karpathy had some context if you want to go through this Sean, I will. But first I'll tell everyone a very important message from CrowdStrike which is super relevant today. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. And I'll also tell everyone about Cisco. Critical infrastructure for the AI era unlocks seamless real time experiences and new value with Cisco. So Andre Karpathy said. New supply chain attack, this time for NPM Axios, the most popular HTTP client library with 300 million weekly downloads. That's a lot. Scanning my system, Andrej Karpathy says he found a use imported from Google Workspace CLI from a few days ago when I was experimenting with Gmail GCAL cli. The installed version luckily resolved to the previous version, the unaffected 1.13.5, but the project dependency is not pinned, meaning that if he did this earlier today the code would have resolved, everything would have updated and he would have been pwned. It is possible to personally defend against these to some extent with local settings, e.g. release age constraints or containers or et cetera. But I Think ultimately the defaults of package management projects, pip, npm, et cetera, have to change so that a single injection, usually luckily fairly temporary in nature due to security scanning, does not spread through users at random and at scale via unpinned dependencies. So very, very crazy, crazy story. Scott Wu said the Devon review caught the Axio supply chain attack for multiple cognition customers before the attack was publicly known. These attacks will be 10x more frequent in the age of AI. It is critical that repo maintainers start using AI for defense as well. Showing one example below where Devin review caught the attack within an hour of its release. Text minorly edited for anonymization. So I was debating this with Tyler earlier. The question is like, how does this update? Diffusion of coding agents, diffusion of Vibe coding. Is this. I was sort of saying, is this bullish for Cursor Windsurf code readers because you would see an organization that said, hey, we were having a great time Vibe coding, but going forward, we have a standard in this organization that we're going to have more humans in the loop. Does this make people be more inclined to put humans deeper into the situation? Tyler's counterpoint. I'll let you explain how you were saying that maybe this is actually bullish for just more token generation, more code gen. Yeah, I mean, clearly, like, there just needs to be more code review, Right? Okay. The package was still seen within seven minutes by an automated system. Right? That's true. So, like, yeah, I think people will just like, there's going to be much more of an emphasis like, okay, use coding agent to write the code. You also use a coding agent to review the code every time. Like right now, that's kind of a thing you do. Maybe later, if you're in a big team, you have code review, but if you're just doing it solo, maybe you don't do as much code review. Right? Yeah. But it just becomes, I imagine, more embedded within the agents. Right. You talk to Codex, you talk to cloud code. Yeah. There's already like every single. I just think it's bullish overall for cybersecurity. Like, I think every cybersecurity company will probably do well. People are on edge already and everyone's even. Even though this type of attack has happened for years, long before, like the popularity of vibe coding, it just feels like there's a bunch of new solutions that are needed. The kind of incumbent cybersecurity players will do well. They're going to release a lot of new products. I think the question that I have is, like, why Seven minutes, right? Why not check it before it's merged in in the first place? Yeah, yeah. Or just like these are machines, so theoretically they can be constantly monitoring versus like. Yeah, I don't know. And the question is, we're going to be digging into this story more over the next next few days, but I'm interested to know, like, it's found in seven minutes. When is it actually rolled back? If you look at 300 million weekly downloads, like clearly there are people that were downloading it at that moment in time. At all seven of those minutes, there's probably like thousands of downloads. If not, you know, tens of thousands just doing like a rough ballpark on what seven minutes means over a week of 300 million per week. But the question is like, how quickly was it rolled back? So is it only if you're in that seven minutes or was it, it was discovered in seven minutes and then it took them another 20 minutes to roll it back and stop serving the contaminated package. Understanding the scope of this, because it's very clear that as Andrej Karpathy explained, like he was actively using it every single day and yet was not caught in that seven minute window. And so he was cleaned. And understanding the scope and scale of the impact is very much determined by how many just, just how, just, just how broad and how many installs happened during the contamination. Anyway, Will Brown has a good take. He says, I hope someone at Axios is reporting on this. And I completely agree. It's gonna be, it's gonna be confusing when they do. Anyway, let me tell you about Gusto. The unified platform for payroll, benefits and hr built to evolve with modern small and medium sized businesses. And let me also tell you about 11 labs. Build intelligent, real time conversational agents. Reimagine human technology interaction with 11 labs. So moving on, more hacks last night, more leaks. What's going on? Last night, POD code, source code was leaked via map file in the NPM registry. There's just a link to. Wait, someone just actually. Do not click a link. If somebody ever says, hey, I got some really great source code here, just click this link. Probably don't click it. Let other people screenshot it. There's plenty of meta analysis over here. Seems messy. Seems unfortunate. Heart goes out to the folks who are dealing with the situation at the same time. Codex is open source. It's not the end of the world. But it did reveal a bunch of things about the roadmap and also some of the eternal April Fools. That is the worst part. We love a secret Surprise. April Fool's joke. I love a good joke. And nothing spoils a joke like hearing about it a day early. But much more importantly, there are lots of other critiques of the way CLAUDE code is implemented. What are the. I don't think this. I don't think this hurts their business at all. No. Because people are using Claude code to make other products. Yeah. And then also having to take basically a fork of CLAUDE code, maintain that, try to be shipping features against it, which is, again, I think it's. It's. It seems to not be legal at all to just fork the code base just because it's out there. Oh, yeah. You can't just. People are converting it. People are converting it into other languages, and maybe there's some argument there, but still, I don't think this hurts their business at all. Understand some of the secrets, what's special, but at the end of the day, all of these tools, especially something like Claude code that's so new, it's more of, like, the process. It's more bad for the overall brand of vibe coding. Totally, totally. Yeah. Yeah. It's rough. And the irony here is that every time Anthropic has released any feature related to cybersecurity, all the big cyber companies have been selling off tens of billions of dollars. Yeah, yeah, yeah. The question of, like, yeah, does this build trust in using vitamins? Overall hurts some trust. But. But again, you know, very obviously going to get through this. Yeah. So the how it started, how it's going is, of course, landing like a ton of bricks in the last 30 days. 100% of the contributions to Claude code were written by Claude code. And the how it's going is that it leaked the source code, which is not what you want to have happen. Yes. Angel says Mythos is so good at security that Claude code source code got leaked. Okay, let's. I don't know. Should I. This is like, you know, you. You didn't get to watch the Super bowl. You have it DVR'd at home. Do you want spoilers? Should we review the April Fool's joke, or should we leave it unspoiled so that we can enjoy it tomorrow? What do you. I mean, it's not. It's not. It's cool. It's very cool. You've already read it. I read through it, but it's not. It's not to my. Were you belly. I don't think we're getting a knee slapper out of it, but it's very cool. Okay. I think it'll be cut. Okay, well, then we can move on. What else did we learn? Tukey summed it up here. Do you understand what just happened at Anthropic? Someone on their team ran a production build of Claude code. The compiler generated a map file, which is literally a blueprint that reverses the entire code base back to its original source. And then they published it straight to NPM for the whole world to download. And it really does show you how fast the NPM downloads. Like there are people that are downloading it every single minute. And so even if it's only up there for a minute, someone's going to get it. And then all they need to do is send it to somebody, zip it, and post a link on X and it goes viral. It's like locking every door in your house, installing cameras, hiring armed guards, then accidentally uploading your floor plans to Google Maps. Does that matter? No, that's a bad analogy. I don't like that analogy, because floor plans are not why I lock every door in my house. I install cameras, I hire armed guards. Aren't floor plans public on like Zillow? Oftentimes they're not. I would say we can probably skip over this. John, if you scroll down, this account just kind of posts like the same format every single time. So we can skip this. Let's go over to. There's a red alert emoji. You got my attention. Let's go over to Lee San. Algae. Yes, yes, yes. A few takeaways from the Claude code leak. Anthropic is actively using Mythos for development. Okay. They are already at Capybara V8. We learned last week that capybaras are extremely deadly, but can be deadly in the right context. Capybara still has issues. The foreshadowing. The foreshadowing is crazy. We were talking about how the Faustian bargain that is getting a capybara as a pet seems so cute, but it can bite you. And it seems like that might be what happened. Capybara has 1 million token context window and fast mode. Numbat is another interesting codename tagged with Odell Launch remove. This section where we launched Numbat. Fennec seems to be the Fennec fox. Fennec Fox is very cute, but also not a domesticated animal. How about we get some golden retriever code names? How about Big Fluffy poodle? That's a good code name for your animal themed AI model. Anyway, let me tell you 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 Lambda Lambda is the super intelligence cloud building AI supercomputers for training and inference that scale from one GPU to hundreds of thousands. Jord says hot take Anthropic leaked Claude code intentionally to get a Nerdosphere code review it would have never gotten if they had just open sourced it. Oh, that's actually true. Way more. You don't leak your entire feature roadmap and you don't do. I mean, it's funny. And I'm sure they'll make the most of this. This is 4D chess right here. But I'm not seeing the 4D chess. I'm seeing the 40 chess. Now I'm convinced. This is. I mean, we're in completely uncharted territory for marketing stunts and pre releases and sneaky footage that is goes viral and maybe was planted. You don't know. And it's like some leaked account. Like, I don't know. I think everything's. I think the gloves are off. Everything's on the table. This could be an April Fool's joke. This could be a stunt to draw to drive attention to an open source move. Although, Tyler, you said that Dario is not a fan of open source at all, right? He's like against, unilaterally. He doesn't want to do open source. I feel like. Isn't there some steel man there? Where, where if you open source, like, I don't know, like, like Opus 2 or something that's like really old. It's entirely commoditized in the research community. So all of those secrets that went into like making Opus 2 good, those have been commoditized. They've been discussed at the house parties in sf. The researchers have moved from one place to another. So everyone knows these, they've implemented, they're available as open source. But by, by open sourcing your model, you can share with more of like the up and coming academic community. Like if, if I'm a. If I'm a computer scientist. Yeah, but if all the research is already commoditized, then yeah, I guess you could just use the other ones. It doesn't really have a benefit. Maybe has any. Has anyone at Anthropic? Has anyone at Anthropic commented on this at all? I haven't. I haven't seen anyone. I haven't seen anyone. Yeah. What is undercover mode? That is a way to contribute to projects without letting people know that you're using Claud code. Oh, interesting. That's a Hack Gurgley over at Pragmatic Engineer says Gergi, sorry, this is either brilliant or scary. Anthropic accidentally leaked the source code of Claude code, which is closed source repo sharing the source are taken down with dmca. But this repo rewrote the code using Python and so it violates no copyright and cannot be taken down. Okay, and there's a warning. Do not store the code even though it has leaked. Do not store it because you might get DMCA'd. According to Primeogen, the last time Anthropic in their infinite PhD level wisdom leaked their own source code, February 25th, that happened. I missed that entirely. They DMCA'd all repos that had their code. Careful storing the code because Anthropic will have no mercy. 40,000 users forked it, so maybe unfork it if you did that because it sounds like you might get a legal letter. Again, a DMCA is not like an actual lawsuit. It's more just unforg, unforgun fork it. And people are of course making a joke that the Codex source code has been leaked in full here and they're linking to the GitHub because codex is open source, which is cool. I don't know. It's interesting. And more and more people are building their own harnesses. There was some interesting data that Opus performs extremely well, better than in clog code, in cursor on some benchmarks. And so there is this new paradigm of like, how can you add different value when you're building a harness? So what else? Yes, certainly there's other. Plenty of other companies that are building harnesses that are going to be able to dig through this and get some benefit, be able to improve their products. That's not. That's not great. But at the same time, Codex has already. The source code is already open source. Yeah. And so that's not. That hasn't been hurting Codex's progress and growth. So, yeah, end of the day, ultimately, I would say, I would, I'm assuming very, very embarrassing for the, for the individual that ultimately contributed to this. But they will get past it. Well, we will put the blame squarely on the AI model so they can take it. Dax is getting line of code mogged. Loc mogged. Because clog code, open source is clogged. Code source is 512 lines of code, whereas open code, his project, is only 118 lines. So he's got to get those numbers up and GT mogs everyone with over 50 billion lines of code. He doesn't have 50 billion lines of code. GT Gary Tan will be coming on the show hopefully this week and we will get the full scoop on how he's using Gstack and other models, which should be fun. Before we move on, let me tell you about TurboPuffer, serverless vector and full text search built from first principles and object storage, fast 10x cheaper and extremely scalable. And let me also tell you about Vibe Co, where D2C brands, B2B startups and AI companies pick channels, advertise on streaming TV, pick channels, target audiences, and measure sales. Just like on Meta. Zach says NDAs are a great way to keep your corporate secrets safe from one or two beers, but not three beers.