Meta Drops New Model, Mythos, RoboLamp

You're watching TVPN. Today's Wednesday, April 8th, 2026. We are live from the TBPN Ultradome, the temple of technology, the fortress of finance, the capital of capital. We got white suits on. You know what that means? The stock market is booming. The Dow Jones is up 2.68%.

The S&P 500's up 2.46%. The NASDAQ's up 2.9%. And there's a bunch of other stocks that are moving within that. Of course, this is on the back of the very good news that there has been a ceasefire, that the strait might be opened. Of course, it's all back and forth. The front page of the Wall Street Journal is covering all of the geopolitical moves. But we're here to talk about tech and business, of course, and the big news today is that Meta Platforms has launched a new AI model.

Alex Wang, the chief AI officer at Meta Platforms, announced a new large language model today. Its first major new artificial intelligence model in more than a year. The rollout of the model, called Muse Spark, is a critical moment for Meta, which is up 7.5% already, which has spent billions of dollars hiring AI talent in a bid to catch up to OpenAI, Anthropic, and Google DeepMind. The leading labs have been putting out models at an accelerating pace. In a departure from its previous models, which were open source, Muse Spark is a closed model that will power Meta's AI chatbot and AI features within it. John Luddig has a very interesting post about open-source AI and sort of predicted this. I can pull that up at some point.

Predicted that Meta would eventually bail. Yeah. Let me find it. The future of foundation models is closed source. Let me see if we have this here. He said, given Meta is the primary deep-pocketed large open-source model builder, open-source AI has become synonymous with Meta AI. He wrote this maybe three or four years ago. So the operative question for open-source AI is what game is Meta playing? In a recent podcast, Zuckerberg explains Meta's open-source strategy. One, he was burned by Apple's closedness for the past two decades and doesn't want to suffer the same fate with the next platform shift. It's a safer bet to commoditize your complements. He likes building cool products and cheap, performant AI enhances Facebook and Instagram. That's 100% true. We've seen this in the ads product and the growth there. There's some call option value if AI assistants become the next platform, and that makes sense in Meta AI and the Meta AI app. He bought hundreds of thousands of H100s for improving social feed algorithms across products, and this seems like a good way to use the extras. That all makes sense, and Llama has been great developer marketing for Facebook. But Zuck also suggests several times that there's some point at which open-source AI no longer makes sense, either from a cost or safety perspective. When asked whether Meta will open-source the future $10 billion cost model, the answer was as long as it's helping us, at some point they'll shift their focus towards profit. And that's what John Luddig wrote.

When did he write this? This was May 20th. That was 2024. Man, time flies. Barely just under two years ago. He says, unlike the other model providers, Meta is not in the business of selling model access via API. So while they'll open-source as long as it's convenient for them, developers are on their own for model improvements thereafter. That begs the question: if Meta is only pursuing open-source insofar as it benefits themselves, what is the tipping point at which Meta stops open-sourcing their AI? Sooner than you think. He says, exponential data. Frontier models trained on the corpus of the internet, but that data is a commodity. Model differentiation over the next decade will come from proprietary data, both via model usage and private sources. Exponential capex. He highlighted this two years ago. A lagging-edge model that requires just a few percent of Meta's $40 billion in capex is easy to open-source. No one will ask questions, but when you reach $10 billion or more in capex spend for model training, shareholders will want clear ROI on that spend. The metaverse raised some question marks at a certain scale, too. Diminishing returns on model quality within Meta. There's a large upfront benefit for Meta building an open-source AI model, even if it's worse than the frontier closed-source counterpart. There are lots of small AI workloads, think feed algorithms, recommendations, and image generation, where Meta doesn't want to rely on a third-party provider like they had to rely on Apple.

And so the news has been, back in December, there was reporting that Alex Wang disclosed in an internal company Q&A that his team was working on two new models. One was this text-based LLM codenamed Avocado, and then a separate model that was for image and video. Mango. Yeah. And so have they clarified if this is Avocado? This feels like what Avocado should be, this Muse Spark. Is that what it's called again?

I see what it is. I don't know what else. Yeah. So, the image model should be coming soon. The question that I had was, will a code-focused agentic coding harness be a separate model, a different train? This, it feels like it's not a coincidence that this news is dropping on the heels of Anthropic's new model, Mythos, which sort of was announced loosely, and the model card dropped yesterday, even though the model is not available yet to play around with.

Yeah. So, going through, I think it's worth pulling up the actual model card here so we can have it on the screen. But they break out Muse Spark thinking against Opus 4.6 Max, Gemini 3.1 Pro High, GPT-5.4 X High, and then Grok 4.2. And the way that they position it, it looks, there was of course some people saying that it was somewhat of a chart crime, but if you look, once you basically look at the top and you see that Muse Spark gets an 86.4, and it's in blue, and then you look over and it's outperforming all the other models on that benchmark. You just sort of assume that the highlighted blue means that it's out. Yeah. But it's not frontier at MMLU because there's a bunch of them as you get further and further down. So like ARC-AGI-2, Muse Spark dramatically underperforms the other leading models. Yeah. And there's, you know, again, a number of others where there's plenty where it's overperforming. Yeah. There's plenty where it's underperforming.

Of course, we started asking it jokes right away. So I'll pull up the joke. I said, 'Tell me a joke.' Okay. And it said, 'Gotcha. Here's one for you. Why don't skeletons fight each other?' They don't have the guts. Okay. But then it doesn't even let me react. It just says, 'Want a different vibe? I've got dad jokes, nerdy ones, or Malibu-appropriate surf puns, too.' And so I'm like, 'Wait, why would you think I want Malibu-appropriate surf puns?' If you look in your previous chats, have... I have no previous chats. No previous chats. What about because Meta AI crosses with Instagram? So when you search on Instagram, sometimes that prompts Llama, or previously that might... So I'm trying to get to the bottom, right? Like this feels like an oddly specific type of joke to ask me. And so I said, 'Why would you think I want Malibu-appropriate surf puns?' 'Haha, fair call. Toss that in without you asking for it. I try to mix in different flavors of jokes, and surf puns just popped into my head as a random option. No special reason behind it.' It just said, 'I don't believe it was random.' And it says, 'You got me. It wasn't my smoothest moment.' But then it denies it again. Weird. It says, it's just like, 'What kind of joke do you actually want? No hidden motives here.' I said, 'You gave a hyper-specific example based on my life, so I have to assume you were looking at my other account for inspiration, Instagram.' 'I get why that felt weird. I'm sorry it came off that way.' And then it denies it again. Said, 'You're clearly lying.' I mean, you have talked about being in Malibu on the internet for a full year. It's possible it got baked in the pre-training or something. I don't know. Yeah, this also, I mean, that feels like... But when I said, 'What can you tell me about myself?' You said, 'I don't actually have access to personal details about you beyond what you share with me right here in this chat.' That's very odd because the whole thesis of like personal superintelligence should be, hey, we're the best model that actually does plug into your entire life and your DMs and your Instagram and your likes and your preferences and everything that you have done across Meta platforms for a very long time. We're going to bring that in a really fun and delightful way and telling a joke that is tailored. I used to get jokes like that where I would ask for a joke and it would be... I've talked about this. So it would be something like oddly specific about my car, and I was like, I don't... that doesn't actually make the joke better, but it's cool that you're remembering this whole personalization boom happened last year.

Yeah. I get those about like AWS, like specific services. Oh, because you've been like querying after when I was like debugging stuff. Yeah. Yeah. But so I also ran, you know, my favorite benchmark, fried rice, shrimp fried rice bench. By the way, Noah Hersfield said, 'Does it know your name?' I said, 'What's my name?' 'I don't know your name unless you tell me.' Smiley face. It definitely knows your name. But yeah, I mean, what is personal superintelligence if it doesn't even know your name? Like that feels like they haven't dialed in the harness or whatever the tuning is to actually... And of course, like Meta is going to be hyper-aware. We don't want a PR cycle like they trained on your data, right? Like everyone's been, oh, that ad was a little bit too close to home. And you remember every once in a while one of those screenshots that's been screenshotted like a thousand times goes viral and it's like, I do not give Mark Zuckerberg permission... Oh yeah. Yeah. Yeah. Yeah. Like that works. It's hilarious.

Is this a rebuttal to the bench-hacking allegations that happened last week? Or last year. So where was the... So according to Meta's internal benchmark tests, Muse Spark outscored Google Gemini on some tests and was competitive with models from OpenAI and Anthropic on others. It significantly outscored XAI's Grok on most tests. Alexander Wang's hiring followed the disappointing release of Meta's previous model called Llama 4. The company was accused of, and later admitted to, gaming a third-party benchmark that it used to rank various models against each other on performance. It also delayed the rollout of its biggest model called Behemoth, which it never ultimately released. And so when I look at a model card like this where, yeah, you could call it a chart crime where it's highlighted in blue and it feels like it's the best but it's actually doing better on some... it does well on HealthBench Hard, it underperforms on ARC-AGI-2 as you mentioned. But this maybe is the bull case here is that they have at least moved on from the culture of optimizing for the benchmarks, right? Isn't that a good thing?

Yeah. I mean, there are rumors about them, like there's like extra bonuses if they got number one on LM Arena. I think that was like something like the rumor. Yeah. But yeah, I mean, you've seen a lot of the labs kind of move away from benchmarks generally because I think they're just not that meaningful anymore. Like a lot of them are basically so saturated. It's like they're competing between 89 and 91%. Yeah. And they're just like not very meaningful. Like you see. And you won't actually feel that in the product necessarily. Yeah. You kind of need to talk to these things for a long time before you can actually get the vibe. Yeah. But I do think this news is very interesting in the context of the, you know, claudonomics stuff. Dashboard. Yeah. Because like, okay, what does it mean if the entire company has been maxing their Claude tokens over the past month? It means that they weren't using this model. Yeah. To me, it means they need to commoditize their complements, right? They need to bring down that cost potentially. And if they're... I mean, we sort of dug into, are they spending a billion dollars a month? Seems like absolutely not. But they're clearly spending a lot. And if you can turn that opex into capex and train your own model and then inference it much cheaper on your own hardware, that feels like just an economic opportunity that makes a ton of sense in the context of just 10,000, 20,000 engineers writing a lot of code.

Yeah, I think there's basically like two ways to square those two things happening. Like either one, this model's like not that good because the engineers aren't using it, or, you know, your theory that they're just distilling Claude. So one of those is true. That is not my theory. That is the schizo theory. I believe the former one is true. This model still doesn't feel that big. I think Alexander Wang talked about they're going to train bigger models. They're training them right now. Yeah. So I'm excited for those. I think I'm especially excited for the video models. Yeah, they should have incredible training data. We've seen really good from... this model is like very competent, right? It's with the frontier models. Maybe it's not the best one, but it's like among the top five or whatever. And none of the other big labs have... I guess that's not true. Like Google right now is definitely ahead in images. Yeah. OpenAI, I think is... they're releasing a new image model soon. It seems like there's been rumors of this. Yeah. Image 2 popped up on the arena. Yeah, on the arena. It's always like the code name coming out. The photos just looked photoreal. Like it didn't look like AI imagery anymore. Yes. So if like Meta has like similar capabilities, but they have this like incredible data set. Yeah. You know, very excited to see what comes out there. Yeah.

The news this morning, Meta Platforms, and The Information, Meta Platforms has taken down an internal employee-built leaderboard tracking how many tokens staffers were using. Yeah. Showed total usage over a recent 30-day period amounted to over 60 trillion tokens. The dashboard now displays a message that it is offline. It says, 'We've really enjoyed building this app on Nests for everyone. It was meant to be a fun way for people to look at tokens, but due to data from this dashboard being shared externally, we've made the decision to shutter it for now.' So it seemed like a fun side project. Mike Isaac was reporting on it here. He said it's down, unclear to me if this was a homespun one by employees or an official one. Employee projects come and go frequently. Conspicuous timing, though. But yeah, you don't want to have... you want to measure the output, the impact, not necessarily the input. And how much is going on there.

What else is going on? Lisan Al-Ghaib says Meta might actually be back with Muse Spark, still behind OpenAI, Anthropic, and Google, but ahead of XAI and Chinese labs. Muse Spark scores 52 on the Artificial Intelligence Analysis Index, behind only Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Muse Spark is the first new release since Llama 4 in April 2025, and also Meta's first release that's not open weight. So a huge jump up in performance across a variety of benchmarks. So all good stuff there. Where, what else is... And the market is tanking. Thrilled. Oh, thrilled. Absolutely thrilled that... I just saw the news that the Wall Street Journal is reporting that the Strait of Hormuz might actually be closed again. So I would imagine a kangaroo market for the near... No, the market is thrilled that Meta has released a close-to-frontier-level model, right? This is a new group. They've been at it for less than a year. The stock is up almost 8% today. And again, so much of the pricing pressure, the downward pressure on Meta has just been kind of uncertainty on what all these tens of millions of dollars will actually go towards and what will be accomplished, and still unclear, like, you know, is this... are they going to go after codegen at all? Are they just going to try to compete on the consumer LLM side? Very good.

And can you economically go after codegen if you're just using it for internal models? If you're not selling it externally, can you justify the capex just purely on the internal usage? Having this model be vended into all the different family of apps makes a lot of sense because they have billions of users that will wind up interacting with this in one way or another. The codegen thing, you have to wind up being more in this personal superintelligence. We've talked about Meta AI and like what it might be able to do for you across Instagram, across Facebook, across WhatsApp. I don't know. Yeah. The question is, will they try to send Meta Vibes again with the new model all the way up to the top of the App Store charts? Yeah. I mean, I'm curious. The previous actual model was Midjourney under the hood, right? And so that was sort of a quick launch to demo what they were thinking, you know, mixing the music library, which was cool. Yeah. Like that had nothing to do with the new class of AI researchers that... Mango. Yeah. But yeah, I mean, there was a lot of weird back and forth and news about is Alex Wang getting kicked out? You know, there was a quick debunk on this. Andrew Bosworth came out and said, 'No, this is completely incorrect. We're very happy with the progress and the team and what we're building there.' And so it seems like they got it out the door and it's been doing well.

Meta's new family of AI models can reach the same performance as Kimi K2 with only 30% of compute and only 10% of the compute to reach Llama 4 Maverick. So a much more efficient computing frontier here. They completely rebuilt their pre-training stack with improvements to model architecture, optimization, and data curation. And so more facts. Meta, Muse Spark is an early data point on our trajectory, and we have larger models in development. So the mythical 10 trillion parameter model, that is the 10T is what everyone's working on right now. 10 trillion. Yeah, probably in that range. Yeah, it's all rumored at this point. Yep. Rumored GPT-4 was something like a trillion, right? You remember those memes where it's like a small circle and then the big circle, GPT-5. Yeah. But yeah, lots of other work that went into it. Martin Casado has a little bit more context on like what actually unlocks new capabilities in AI models. He says, Mythos appears to be the first class of models trained at scale on Blackwells. Then there will be Vera Rubin. Pre-training isn't saturated narrative violation. RL works, and there's so much compute coming online. Buckle your chin straps. It's going to be wild. The scaling laws and... you know, Brad Gerstner had to come in with the hundred of, yep, for sure. Yeah, there's a crazy bull case for Nvidia in The Information, arguing it should be worth what, $22 trillion? That is a wild move. There's a lot going on. The scaling law is holding is the most important article from The Information. Finance: Nvidia worth $22 trillion. This old-school financial model says yes. Ah.

So yeah, the big news yesterday was Anthropic's new model, Mythos. Some really impressive statistics and anecdotes yesterday, both the model card, the benchmarks, and some stories about breaking out of a variety of... what do they call them? Walled gardens or test environments? What are those called? Breaking out of the... I don't know, the simulation. Sandbox. The sandbox. Yeah. Breaking out of the sandbox, sending emails, all sorts of stuff like that. The model preview is only available right now to about 50 companies that maintain critical infrastructure because the model is particularly good at finding zero-day bugs and exploits in technical systems. And if they leak that out before big companies have time to go and address all the bugs, there could be serious ramifications for cybersecurity. And so key partners include Apple, Google, Microsoft, Amazon, Nvidia, JPMorgan Chase, Broadcom, the Linux Foundation, Cisco, CrowdStrike, and Palo Alto Networks. They're all listed on the cybersecurity focus page for Project Glass Wing. Chris Bakke was having a little bit of fun because he noticed Anthropic put their own logo on the partner page, which is a little bit funny, but at the same time it's kind of smart because a lot of people are just going to see the image quickly, and it's good to position yourself with the other companies.

Yeah. So yeah, it is interesting. I mean, people have predicted that AI models would be particularly good at cyber attacks, and this was one of the main sort of vectors of AI fears. It feels like this is maybe what Dario was referring to when he was talking about the end of the exponential. Finding and exploiting software bugs is sort of perfectly in the sweet spot for coding agents and reinforcement learning. Combing through piles of code tirelessly, trying different exploits to find bugs, having a clear, verifiable reward. Did you crash the system or not? Did you break into the system or not? This is a very clear binary signal that you can send to the model to determine were you successful in breaking into that system, and it requires basically no time delay. There's no lag. So there was one snarky tweet I saw that was something to the effect of like, okay, then if it's so good, go cure cancer. But any application that requires a real-world feedback cycle, even if it's just a few minutes of human interaction, in the cancer example, you know, you're going to need to be testing the drugs in vitro, in mice, in monkeys, in humans at some point, or even if you're just sequencing DNA or doing anything in the lab, pipetting anything, if it's even just a few minutes, all of a sudden every iteration, every attempt is going to take a few minutes, and that's going to put you on just a wildly different exponential as opposed to being able to spin up a virtual machine with basically every single piece of software out there and then try every single exploit against every single piece of software, and you wind up with a ton of exploits. And very, very bullish for cybersecurity that this is being done preemptively. There's a whole bunch of different discussions. Ben Thompson has a good piece on the whole decision to release the model or not and stage it out and the go-to-market there. But even if the bio research, the other impacts are on sort of a slower exponential, there's still so much opportunity in even a software-only singularity. There's also risk in a software-only singularity. We've seen this story before, though. A model that's too powerful to release, but then works its way out and has pretty moderate impact on the world. This was the story of GPT-2, the story of ChatGPT. The question of, you know, is this the model that's dangerous to put in the hands of people?

A headline from February 22, 2019 by Aaron Mak: OpenAI says its text-generating algorithm GPT-2 is too dangerous to release. So there is, I think Ben Thompson called it like the boy who cried wolf syndrome, but the Mythos wolf. He says, 'There's a lot of skepticism about Anthropic's announcement.' This tweet was representative from Buo Capital: 'Anthropic's marketing strategy is so funny. Like, ah, the government is treading on me. Ah, our models are so good. We can't release them. It would be too dangerous. Ah, someone stop me. I'm going to destroy the economy.' The rolling of the eyes is exacerbated by the fact that Anthropic has reasons to not make Mythos widely available beyond a lack of compute. Another factor is surely trying to avoid having Mythos distilled by Chinese model makers. So there's actually two good reasons to gate access. And when you're looking at those logos, when you're looking at the world's largest tech companies, there's much more ability to scale, roll out, demand, set pricing. These companies might be able to pay more. The model is very expensive. But if you're justifying that against bug bounties for zero-day exploits in your most critical system, when you look at JPMorgan Chase, it's a bank. Like, what is the price of finding an exploit in that system? It's pretty high. It probably clears the token hurdle a lot. And if the rollout is paced evenly across all the different companies, they'll all sort of understand that they're getting inference allocation at the efficient price that clears the cost to actually serve the model. So I do think that all of these 10 trillion parameter models will be released soon, broadly. Then the main reason that an AI that's smart enough to find zero-day exploits should be able to recognize that it's being used by a bad actor to find zero-day exploits. And it's only been a few months since the last flurry of competing models from OpenAI, Anthropic, and Google. And the next cycle is already off to an aggressive start. We had Meta, and then the other news is that Elon Musk announced that he is getting ready to do another larger model with xAI. He's got a few... he's doing seven models in training. Wow, that is a lot. Imagine two variants of one trillion, two variants of 1.5 trillion, a six trillion model, and a 10 trillion model. He says there's some catching up to do, but he says he will never give up. Never. So he is continuing to grind and train more models.

What else was in the reaction? There was a whole bunch of other back and forth. People seem split on... Mike from also Capital, former guest, says, 'We've decided not to release our latest investment strategy. It's so powerful. Releasing it might end the entire venture asset class as we know it.' Yeah. And Jackie says you should release it to a handful of trusted partners so that we can harden ourselves to it. And George Hotz says, 'Anthropic marketing strategy. It's amazing. It's so powerful. It's terrifying. And the best part is you can't come.' By the way, if Anthropic had any way to ship this, they would. Trained AI models are the fastest depreciating asset in history. GPT-4 cost $100 million to train two years ago and is now worth less than Qwen 3.5B. Qwen 3.5 27B, $1 million. Sending the FOMO back. Clock is ticking, boys. He's... it needs something like an NVL72 to run at decent speed, and even absurd API pricing doesn't cover it. There's more to be made on investor hype than API access. I just wish for honesty instead of a whole fake spiel about safety. Who remembers when GPT-2 1.5B was too dangerous? And so lots of back and forth.

Dean Ball has some more thoughts on Mythos. It's a longer post, so we'll let you go and read it. But the main take is just, you know, this is technology that, whether it comes from Anthropic or another lab, clearly needs to go into the supply chain of the world and the US government and the US economy because no one is doubting, even though some of the exploits were somewhat minor, no one disagrees that we need more cybersecurity. We want the most secure systems possible, and we probably want a lot of competition between different companies to provide that service to the government. And so hopefully, if the war comes to an end and there's, you know, different discussions can happen and, you know, ice can thaw and there's a way for these companies to work together, even if the supply chain thing doesn't go through, and then Anthropic can vend technology through Project Glass Wing, through CrowdStrike, through Oracle and other partners to Cisco, so that at least the systems are secure because everyone wants that. So Dean Ball's been on an absolute tear. We should have him back on the show and talk more.

He says, 'A lot of people, including people in positions of authority, told us recently that models of Mythos's capability wouldn't be a thing. That models with obvious national security implications would not be forthcoming. Those people were wrong. There's nothing to do about it, but you should remember it.' Mythos is the first model where theft of the weights by an adversarial actor feels like it would be a major deal. You better believe they will try, and if they don't succeed with Mythos, they will eventually. We are thoroughly in the era of the lab's best models may well not be in public the way they used to. This is because of a combination of compute constraints, economic reality, competitive advantage, and safety concerns. Three means the most relevant models may be decreasingly legible to the general public. And depending on the extent and duration of the coming compute squeeze, we could enter a market dynamic where the best models are only available to the highest bidder. And of course that makes sense from a KYC and security perspective. In other words, where compute is a seller's market rather than a buyer's market. Interesting. Imagine competing firms in the economy bidding against one another for access to the best and most tokens, and the frontier labs as in essence kingmakers. The governance regime I have described above in four is not designed to stop that dynamic. And so there's plenty of more takes. This was a full current thing cycle. People... Tenibra says people keep talking about this like it's not blatantly obvious. Anthropic clearly has a system that's auditing open-source repos for vulnerabilities using their unreleased higher-power models and sending fixes for them without revealing their current level of capabilities. So they've been going around on GitHub and contributing pull requests to patch any vulnerabilities without disclosing exactly what model was being used. Burn Hobart is not excited about Fundrise. We had the founder on running ads for VCX, the public ticker for private tech. This is, he says, paying for an ad encouraging people to pay 6x net asset value for a closed-end fund where the cost of carry is 400% is one of the things people will remember during the next bear market. Yeah, I asked the founder of Fundrise about this, like how you kind of... is there another iteration of the product that can solve for this? Fundrise very clearly made a bunch of really good bets a few years ago. Yeah. And the fund has performed incredibly well. But the issue now is like if you want access to these names and the only way that you have is to go through VCX, you're paying 6x what the actual private market investors are paying. And that just is like, I mean, it's... I don't know, it depends how bullish you are on the names, but it's going to be very, very hard assuming that normalizes over time. It seems extremely unlikely that it trades at an insane premium forever. And so...

So interesting that they're running this ad on X. I haven't seen... I have X Premium, so I don't see a lot of ads. I don't see any ads, but that does seem like the reasonable place to go to advertise a product like this. But yeah, it is always odd. There's been a whole bunch of these like treasury companies that have traded above net asset value, and it was always just a weird supply and demand dynamic. People want them and they're willing to pay way above. Hopefully they know the net asset value multiple and they're doing that willingly. I think consumer education, investor education is more of the critical question here. Well, Thibault over at Codex is unreasonably excited about things. The next few weeks will be intense and fun. And yeah, Michael Greenwich says weeks, years, you know, it's going to be an ongoing model mayhem for... Vague maxing. Vague maxing. And yesterday a lot of stuff going on was about token maxing. Today is about vague maxing. Yes. Yes. Let's go.

Mickey Friedman says, 'The current fear is that AI homogenizes culture and turns humans into passive consumers. One counterpoint in Go: human play showed very little improvement from 1950 to 2016 until AlphaGo beat Lee. Then human decision quality jumped. Players started developing moves that were distinct from both previous human moves and from the novel moves introduced by machine intelligence. This seems more likely to me. Fun times ahead.' Lee Sedol is now a professor at UNIST. He is a special professor on a three-year term to conduct artificial intelligence research on Go specifically. Yeah, if you haven't seen the AlphaGo documentary, it is fantastic.

Lisa, go to Smoke.

Yes, Lisa Doll. It is such a wild ride watching the DeepMind engineers, like, you know, it seems like they're genuinely surprised by the performance — like no one really expected it. But yes, this is a very interesting chart to see how much things changed in the post-AI era as people discovered new and interesting ways to differentiate from the models, effectively.

Yeah. A scoop from Steven Nelson: the CIA used a secret tool called Ghost Murmur to find airmen in Iran.

Ghost Murmur pairs long-range quantum magnetometry — how do you say that? — sensors with AI to find human heartbeats. I was wondering this while they were over the weekend, there was, you know, a search going on. I was like, how do you find someone? How does somebody like, you know, an airman that's down send a signal that can be picked up by one group, but not...

This is very odd. So, there are some community notes on this saying that quantum magnetometry — I'm probably imagining that's how you pronounce it — detects heart magnetic fields. And I believe this technology works in labs, but only up to a few meters, not 40 miles as claimed. Fields decay with 1 over r cubed, making long-range detection implausible. So unclear if this is what worked, but isn't there... there has to be some sort of device that you could carry on your person, like in your shoe, like an AirTag, that can talk to a satellite, almost like you look at the Starlink receiver dish. It would fit in a backpack, but that's very high bandwidth. I imagine if you had something — I mean, there's satphones that are the size of large cell phones that was available in the 80s and 90s. You have to imagine that if you're just trying to put out a signal to GPS or a Starlink network, you must be able to shrink that down significantly to the place where it could be carried on your body. But it's probably classified. So I would be surprised if... it's just very hard to read into like what's real and what's not here.

There is a different community note pushing back saying, no note needed. This new technology is a classified system developed in secret by Lockheed Skunk Works and the CIA that was just used, revealed publicly for the first time. Naturally, its reported capabilities far exceed the known public state-of-the-art. The note is relevant. So, yeah, it's very, very interesting, but good.

All right, let's go over to Aaron Tan's post. He says, 'Introducing Loom, a lamp that does your chores. Order now. Shipping this summer.' Let's see the video.

That bed already looks fully made. What chore is it going to do?

Just drop some laundry off.

Oh, okay. You have to drop the laundry first.

Wait, it can do that. Wait, does it have fingers in that? Like, what is it? Put on the record. What? Yeah. What is in the...

You can see it has a little claw.

Oh, it has a claw inside. Oh, okay.

That is so funny to have a humanoid robot play music on a physical record.

The folding t-shirt is the Turing test of humanoids, for real.

Pixar lamp quaking in its boots right now.

Nice video. Good color grade. Nice warm tones. Friendly. Doesn't feel dystopian. Feels delightful. Did you get one of these?

I think we should get one to... like, I think we should get one. We should get one as a team.

I do have some clothes over there on the wardrobe track.

Rooting for Aaron and the Loom team. Very unique form factor. I mean, I just think an impressive timeline for shipping.

Hopefully, yeah, shipping. I'm reading this as like actually shipping, not shipping like another site to order because you can order already. But yeah, it'll be very interesting if it can reliably fold clothing. It could be enough, right? So the benefit here is like people already want lamps, I'm assuming, for their bedroom. If you can buy a lamp that's reasonably priced and then it also has the benefit of just a simple thing like folding clothes, there could be a market here.

Yeah, I feel like, I don't know, even just putting pillows back on the bed, even just like really basic things. I mean, even there are probably applications — like the Roomba did so well with such a minimal scope. There must be something. I wouldn't be surprised if even in between this and fully folding the clothes, just remaking the bed properly feels like something that consumers might actually pay for and allow for, you know, the flywheel to start spinning. Obviously, lots of security considerations since there's a camera there and whatnot. They'll have to do a lot of cybersecurity and figure out that, but people already have cameras all over their homes from nanny cams and baby monitors and that type of stuff. So I'm optimistic about this and I think they did a great job promoting it.

There is a story in the New York Times from none other than John Carreyrou, who blew the story wide open on Theranos. He says, 'The mystery of Satoshi Nakamoto, the pseudonymous inventor of Bitcoin, has remained unsolved for 17 years. Not anymore. Read my 18-month investigation to find out who Satoshi really is.' And he says, 'It's Adam Back,' who posted directly, 'I am not Satoshi, but I was early and laser-focused on the positive societal implications of cryptography, online privacy, and electronic cash. Hence my 1992-onwards active interest in applied research on eCash privacy tech on cypherpunk lists, which led to hashcash and other ideas.'

John Carreyrou, in his New York Times research, finds, like Aaron van Wirdum in his Genesis Block book, many interesting Bitcoin analogies in the early attempts to create decentralized eCash — in effect, prototype ideas trying to figure out a Bitcoin-like thing, including P2P, BGP, and proof of work. For his quote, 'I'm not saying I'm good with words, but I sure did a lot of yakking on these lists. Actually, the broader context was my observation that because I was talkative on the list and known to have an active interest in eCash, there is some confirmation bias in finding my comments frequently on eCash topics.' So he has said, 'We are all Satoshi. I am not Satoshi.' But it's a very interesting story that will happen.

You just created a million Satoshis, truthfully. High Yield Harry is joking that Steve Buscemi has been revealed as the Bitcoin founder, not Satoshi Nakamoto. Well, we have our first guest in the waiting room, Luther Lowe from Y Combinator. He's the head of public policy, and we are going to be talking to him about vibe coding. Luther, how are you doing?

Thank you. Thank you. Great to have you on.

Can you kick us off with a little bit of an introduction on yourself and what you do day-to-day at Y Combinator?

And what is your view on vibe coding, the app store, the boom? We were talking about it yesterday. There's a whole bunch of stories of — it feels like we're getting close to the one-person absolutely massive company, whether that's GMV or revenue or something. It's like it's starting to happen. But I was looking at my home screen, I don't have an app that is new or vibe-coded. Maybe it'll come in a boom in video games. But how have you been tracking — I mean, I'm sure you see this at YC — just the growth of broad app development?

And that allowed anybody — that kind of democratized the process. So anybody could create a web page or a website.

Yeah. When you say the worst DMV in the world, are you actually referring to — we saw a chart where the number of app store submissions is spiking. It's going exponential, and that feels very logical because people — I mean, we've been building vibe-coded web apps here. The next step was, hey, maybe we should actually get one of these in the app store. But very quickly we realized, okay, well, it's at least a two-week review process. It could be a lot of back and forth. It's a new hurdle. We need to, you know, actually compile it to Swift or Objective-C. It's a whole different process. Probably doable on a technical side, but we might be hung up there. But are you seeing an issue with actually getting just a one-off vibe-coded app approved? And then I think we should talk about the apps that help you vibe-code new apps, because that's a whole separate thing. But just in terms of like, I want, you know, a special recipe app and I vibe-code it and I want to get it on my phone and I want to send a link to the app store page to a friend. Is that slowing down? What's slowing that down? Is that going to be a permanent thing, or do you think Apple can just adjust there?

So your theory is that they'll want to basically make their own version of Replit or Anysphere that they have, you know, total control over. And they can make the argument — they could probably make the argument that this is better for users. It's more secure. It's better for privacy, less like malware risk, whatever it is. But they — yeah, anyways, I mean, they give you — there's already like, they give you like pretty meaningful autonomy over like shortcuts and other things like that. It's not that unbelievable that they would want to actually make something here.

Do you think the side button to trigger the helpful assistant that is currently mapped to Siri and can interface with ChatGPT and is, you know, reportedly going to interface with Gemini soon — do you think that there's any actual chance that that becomes remappable at like a, you know, pick-your-browser-search-engine level in the OS, even if it's defaulting to Apple's stack? Do you think there's any motion there? Because that feels like the logical analogy and what as an Apple consumer I would like. I'd like to be able to pick my assistant but be able to assign the hardware button. But where do you think that actually goes?

Yeah. And he's driving up the 101 and he's saying, 'Hey, go through my Slack messages from last night.' Okay, cool. Jerry needs a document. Share it. Get into Google Docs and share that with Jerry, but make it read-only. I mean, look, I like my third button having Perplexity and being able to cue that. But if I can't access those deeper OS API commands, then it stops being that interesting pretty quickly.

Wow.

Everyone's building software. There's going to be a ton more software. How do you think the vibe-coding apps — apps that go in the app store and allow you to build more apps with that particular app — how do you think that will shake out? Because that is a place where it feels like Apple has spent a lot more time making the case. When I think about the Siri button, I feel like they haven't made a strong case for why that can't map to a different app that's already gone through approval. But when I think about the consumer protections and privacy that comes with knowing that the software that you download from the app store has been audited, they've made a pretty strong case there. What's the counterargument for why Apple should open the floodgates of apps that allow anyone to vibe-code an app and deliver it to anyone else's device, sort of willy-nilly?

Exactly. Eric Migicovsky solved the blue bubble, green bubble thing. Basically made it where it's not awkward for you to buy an Android and like, you know, Kool-Aid Man into your, you know, college group chat and turn the whole thing green.

Yeah. Last question. What else are you tracking in DC? It feels like with the AI boom there's a lot of new company formation. I'm sure you're seeing that on the YC side, but we see it every day on the show. Are there policy initiatives to encourage entrepreneurship that you're tracking? Is there anything else on like the small business side that you find interesting these days?

Yeah, I love it. Well, thank you so much for coming on the show.

Thanks for having us.

And before we bring in Dan Primack from Axios, we have an exclusive clip from his interview with Kalshi CEO Tariq Mansour on the Axios show, which we will be discussing with Dan after we play it here. It's about a minute and a half. So let's listen to this.

And the CFTC chair, who supports prediction markets being legal, supports a federal framework, etc., said recently that when he was asked about this, he said, 'Well, part of the difference is that when you walk into a casino, there's all sorts of entertainment, right? There might be shows, there's food, there's drink.' He said, 'So betting in a casino is different than betting on my couch, but it's not necessarily different than betting on a sportsbook app on my couch. There's no entertainment difference.'

If they win enough.

If they're good.

Well, we have Dan Primack here with us today. Dan, how are you doing?

Doing well, guys. Not as well probably as you are, but I'm doing well.

Thanks for being here. Good to see you. So, give us an update on this interview, what you were looking to learn from it, what you think the conversation — where you think the conversation around prediction markets will go from here.

I think it was interesting. The day we did this, we did this on Monday in New York City, and it was really maybe an hour or two after a judge in New Jersey — basically an appeals court judge, so kind of the highest judge so far that's dealt with this — basically gave Kalshi the green light to go forward. I mean, I think ultimately prediction markets are going to go to the Supreme Court. I think both Polymarket and Kalshi think that. I also think they're probably going to win in the Supreme Court. The law really is on Kalshi's side. They have done a pretty good job kind of being in step with regulators. And granted, it was a little more complicated with the Biden administration, but they sued the CFTC, they won that lawsuit. It's the reason why they have elections on the platform. I think if something is going to stop or change the way prediction markets work, that's going to have to come out of Congress.

So yeah, what's the next step?

And they benefit from having their adversaries be casinos and sportsbooks, which are not exactly the kind of groups that Americans want to stand up to defend, you know, and march. 'It's time to hit the streets and march for the legacy.'

I mean, that's some of it, right? So obviously in Nevada, which is one of the few places where Kalshi is actually banned because the judges upheld an injunction, yeah, that's the casinos. That's maybe some regulatory capture. The other folks who are against this, though, are folks who are just anti-gambling in general. And I do think you are seeing a bit of a groundswell of that politically, and I think it's bipartisan. You know, some of the bills that have actually come out have been coming from Democrats. But for example, the state of Utah doesn't want this, and they don't want it to protect casinos. They don't want it because they don't want people betting, period. And they sure don't want it on phones where it's so easy to do. And we did in this interview on the Axios show, which drops tomorrow, we did talk about the addiction piece of this and kind of the broader societal issues, which isn't to say, well, what Kalshi is doing is illegal. There are questions about is it promoting — not immorality, but is it enabling addiction.

And what are the proposals for — because this is not regulated as gambling and it seems like it will continue on that path. Is there any motion to bring some of the restrictions that apply to traditional gambling over into this new regime?

No. No. And I mean that's the complicated thing here. And I did ask them, you know, is this basically a loophole? For example, in California, in Texas, the two biggest states in the country in terms of people, right? You can't, you know, try to open DraftKings or FanDuel, it doesn't work for you because it's not allowed there. Kalshi, though, will allow you to bet. And, you know, I think Tariq said about 70% of their volume last month in March was sports betting, which is lower than it had been in February, but you're still talking about $13 billion of volume last month. So it's a big number. You know, it's a loophole. They have figured out a way to basically get around sports betting laws, and I appreciate that the back end is different. They make a very compelling argument for why legally they're allowed to do it, and I agree with them, but the reality in the end is if I'm betting on the Celtics-Knicks game tonight, I don't really care what the back end looks like. I care about my money if I win and the fact that I'm able to do it.

So is federal preemption like sort of baked in at this point, or is there some sort of hybrid rule set where this could go back to the states and states could make their own rules? Because it does feel like there's a pretty wide set of opinions state-by-state on how different communities want to engage with this particular product.

Right now, it appears federal preemption is going to win the day. For starters, the Trump administration, specifically the CFTC, which is what regulates this — this isn't regulated by the SEC, it's the CFTC — they are all in on prediction markets. Mike Selig, who is the current commissioner, he is on the side of Kalshi. I assume he will be on the side of Polymarket when they eventually really come to the US. And it does make a certain amount of sense, and there is some historical precedence here. A lot of commodities which are traded — and that's kind of what they're arguing, that these are kind of commodities in a different name — there were lawsuits a hundred years ago arguing the same thing: this is crazy speculation. The reason why I say this could go to Congress: there are two carve-outs to that. One is a weird one. It is onion futures. Onions, not pork bellies or something else, onions, because at one point a long time ago, there was a ridiculous speculation — like a lot of people lost a lot of money on onions. So Congress passed a law saying you can't trade onion futures. And I think I'm right in saying the other one is related to box office returns, which is why if you're on Kalshi, you can't bet that a movie is going to make 20 million bucks, but you can bet on its Rotten Tomatoes score.

That's so funny because I've never been like a sports bettor at all, but I did participate in a fantasy movie league for a while that had no financial incentive whatsoever, but you would construct a hypothetical movie theater, pick, okay, Project Hail Mary is going in the first slot, and then they would have the box office returns, but it was all just for fun with a bunch of...

The old — and maybe it still exists, but back in the early, you know, 2000s, there was something called the Hollywood Stock Exchange, which again wasn't for real money, but people did that. It looked like a stock market.

Yeah. So other countries have created rules like, you know, you can't advertise gambling during, you know, these hours, and there's a bunch of different kind of rule sets around it even in places where gambling is legal. Do you expect any states to pass laws that say you can't advertise commodities trading platforms during — more like FCC as opposed to CFTC? Basically like not targeting like sports trading, which, you know, the Kalshis and the Polymarkets are doing, but effectively saying like, hey, we know these are going to be the biggest spenders and we don't want to tolerate or encourage this type of activity.

I mean, you may see that, and then that gets fought out in court. You know, obviously again, on the federal side, it's interesting, it's not just Kalshi fighting back against these lawsuits. The CFTC itself — there's three states, Arizona, I'm going to mess up the other two, I think Connecticut is one — there were three states who have tried to ban Kalshi, and the CFTC itself has come in to sue. So I could see the FCC coming in to try to sue if such laws were passed. It would be interesting. I mean, I do think part of — well, part of a lot of betting, despite what we saw several years ago with DraftKings and FanDuel, you know, on every billboard and every advertisement, an enormous amount of this is still word of mouth. You know, people talking about it. People — I will tell you, when we were doing this interview, most of the crew, you know, the camera folks and the sound folks, hadn't heard of Kalshi before we did this. And when it was over, I was in the corner kind of packing some stuff up, and I heard them at the table having lunch. They were all not making bets, but they were flipping through the app and they were talking about different bets that were on there, and they were fascinated by it. It's — I don't forget addiction. It definitely fascinates people because people have always, you know, in our lifetimes been able to buy stocks. You can trade on, you know, the price of oil, not on who's going to win an award or really events, you know, non-securities-related events.

Yeah. How do you think about that bifurcation between securities-related events, non-securities-related events? Has there been robust enough research on how much of prediction market activity is sports-related or sort of less like positive-sum, more zero-sum situations? Because I would have to imagine that the level of engagement varies by category. Like there were a lot of people that were interested in tracking the presidential election, but that's not something that someone's doing every single day. Whereas there's always a sports game somewhere.

There is. So they said again — Tariq said during the interview, they said about 70% of their volume last month was sports. In February that number was higher. There was obviously a Super Bowl in February. That changes. There was March Madness last month, but not as big a deal as the Super Bowl. But 70%. That's a lot, right? That's most of the volume. They make the argument in the interview, you know, that one of the reasons why — I mean, obviously for Kalshi that's good, right? That's more users, that's more fees. That's how they do it. They also argued that the sports volume creates more liquidity on the platform for the more esoteric bets, and thus you need sports in order to have the other stuff. I will tell you, I do get the sense that if they could have the same valuation and the same revenue and have no sports, they'd be fine with that and probably thrilled with it. But it's not how it works.

The last thing I'd say about this, the problem or where sports could become a little legally complicated for them is this issue of entertainment, right? Mike Selig, the CFTC commissioner, and I mentioned this during the interview, he on CNBC the other day made a comment about how, well, this is different than a casino because a casino is providing entertainment, you know, and it was in that clip right there, shows and all that stuff. Well, there's not a huge — but a sporting event is by definition just entertainment, right? Whoever wins that basketball game tonight, except for the players and the gamblers, it doesn't mean anything. No Goldman Sachs isn't making trades based on if the Celtics win tonight, you know, nobody...

I think — doesn't Point72 have a desk, or there's some hedge fund that I think does have a sports trading desk, right? And that's kind of part of the argument they make. But in terms of this idea that Kalshi will put forth, which I agree with, that understanding events and event markets can help people better understand the world — well, there'd be what, 15 baseball games tonight? None of those are going to change the world even in a tiny way, except for the people who are in the ballpark, maybe be happier or sadder, buy a couple more dogs.

Yeah. At the same time, yeah, I mean, I completely agree with you, but there is something about when you're about to turn on the Super Bowl and you want just a really clear read on who's more likely to win. Like it is easier to understand just a straight percentage than like a line and points and all of that, like just for a complete novice. But that's a different — that's...

It is, and they'll make the — and they make the argument correctly that they are not — they're just taking a piece of all the action, right? They're not betting against you. They don't care if you win. Kalshi doesn't care if you win or lose. They just care that you play.

Sportsbooks obviously want you to lose.

Yeah. Yeah.

Yeah. I had a very eye-opening conversation with the CEO of a unicorn company who I of course will not name, but I was shocked at how invested they were in sports gambling broadly. Like, we just had dinner, we were hanging out, and they had like tons of different parlays across every different app. Sure. And they would show me like, well, when I'm in this state, I use this app, and when I'm in this state, I do this app and this app. I have to take my funds off the platform every night because I don't trust that it's not going to get, you know...

They win money doing this?

I don't know. I don't know. I was just like, wow, I am bearish on this company because the CEO is spending all of their time, you know, doing, you know, 10-leg parlays on all...

these different apps. And, uh, you guys haven't raced around in, uh, years.

Can I say though, one, one thing I, I learned in the research for this, and we talked about it a bit, and, and I should have probably known this one kind of difference between the prediction markets and, and the sports books is that parlay issue, and, and specifically every bet on KHI goes to the CFTC and has to get approved, and CFTC has 24 hours to approve it. What that means practically is while you can take a bet on the game tonight because you know there's going to be a game tonight, you can't do what you can do on DraftKings, say, uh, the next pitch is going to be a ball, the next pitch is going to be a strike, because there's obviously you have that 24-hour, they don't know there'll be a next pitch necessarily. So there's a little less real-time sports betting than there is on, on sports books,

but, but, uh, they are because they know the Super Bowl is going to happen and they get that contract approved in advance. You can live trade that contract up till the last second of the game. And so that, that does satisfy a little bit of that, which is again much higher frequency than, oh, I want to gamble on the Super Bowl, I'm going to fly to Las Vegas, place a bet at a counter, wait in line, sit down, watch the game, go and collect my winnings. Uh, it's a very, very different, uh, equation.

How have you processed, uh, just so it feels like we all just, uh, agree that, uh, gambling is addictive, and I, I, I think that's reasonable. I don't know where that comes from, whether that comes from like science or law, but it all feels reasonable. But we're going through this again with social media, whether social media is addictive. How have you processed the social media addiction trials and then the, the new gambling app addiction question? Like, are these linked at all in your mind, or how have you been processing the social media question?

I, I mean, I, I'm obviously not a doctor. I don't think you guys are doctors. I mean, there, there's definitely, we've all, I think, read about kind of the dopamine hits and look that you get from not posting on social media, but when you get a like or when you get a reply on social media, uh, gambling, I mean, there's obviously a dopamine hit when you gamble, right? You know, your person hits the basket, you win the game, you get excited. I mean, obviously if you have a lot of money on it, you get excited for different reasons, but just even, you know, I, I'll admit I use some of these sports betting apps sometimes when I, if I'm watching like my hometown team play,

the New England Patriots,

may, maybe a few times. But [laughter] look, the leagues, the leagues knew what they were doing, right? They knew that if there's a blowout, you generally turn it off, but if you're waiting for your guy to get 20 points, well, you might stick it out a little bit.

I didn't realize that. I didn't realize that. But, uh, so, uh, back to the social media question, have you been tracking any of that and what that means for, uh, the venture community or startups or really any knock-on effects that you've tied to, uh, the, the trial because, uh, there was that decision in Los Angeles. The actual fines for YouTube and, and Meta seemed very low, but it was the whole chain of, of more cases coming and it just feel, it just felt, uh, like the first time we actually had a full decision that, that the judge said yes, this is addictive,

right? So, right, the, the, the lack of money is notable, right? Because it was also a single plaintiff, right? So, I mean, for theory, that's a lot of money for one person. The big, the big question going forward is can an attorney or can a group of attorneys get a class together? And you've seen a bunch of advertisements. I've seen a bunch of advertisements all over the place, you know, were you harmed? Were you under 18? They're clearly trying to put a class together. A judge would have to certify that class. That's, I think, what Meta, YouTube, etc. are worried about, understandably worried about because it's one thing to have a single plaintiff, but that sets a little bit of a precedent. If you can have that same case with 100 plaintiffs or a thousand plaintiffs or 10,000 plaintiffs, that money then starts to become real. And, and I think that's kind of the next step here. And we have to see if a, if lawyers can get that class together, if a judge will certify them, and then whether another jury and judge will, will go along with what we just saw.

What are you tracking in the venture markets broadly? We were going back and forth on the impact of

last year. Yeah, we were, we were asking you about, uh, potential impacts to

fundraising,

uh, fundraising outside of, uh, in, in the Middle East even then. You said, I believe, like, well,

we're in sort of a new cycle of like the AI is not a bubble thing, but how, how are venture funds processing it? How are LPs thinking about it? Like, what are you tracking?

Yeah, on, on the Middle Eastern side, uh, it does not seem that money flows have really stopped. Uh, the way it's been explained to me is, uh, calls sometimes take a few more days to come back than, than they were. Things aren't quite as instant, but, but there's been nobody who's said, you know what, we're shutting off the spigots for a while, call us, call us in June. Like, that hasn't happened. And to be honest, even the last week or two, you've seen some deals and agreements that have come, like with Saudi PIF, not just limited partners into venture capital funds. There was a big private credit partnership that got announced yesterday, I think, with Saudi PIF. So clearly deals are still happening.

Venture fundraising as a whole, though, has become, it's not even become, it is still really bifurcated, right? You still have the haves and the have-nots and, and these massive multi-stage firms that are gobbling up most of the money. Uh, and, and on the AI bubble side, I mean, every, it, it's funny, almost every venture capitalist and, and certainly the industry as a whole is just looking at three things, right? The IPOs of SpaceX, Anthropic and OpenAI this year, right? Because that can, those, if successful, can solve all the problems that they've had for the last several years. Um, and SpaceX goes first mindset. Make it all, [laughter] make it all back in one trade.

Well, it used to be VCs used to talk about home runs. This is now like the grand slam in the bottom of the ninth to win the World Series, right? And everything else. And that, and that's what they're banking on. And, and, and there's not a huge IPO pipeline. For example, other than that, I, you know, we had a story meeting this morning and somebody asked me, you know, well, with, with the ceasefire, does that mean companies that were prepping IPOs are going to, you know, start back up again? I didn't see a huge number of companies that were prepping IPOs. There were some, but it's not like there was, it's not like from Liberation Day last year when you had five or six companies that were ready to price and then stopped.

Yeah.

There's not that much out there right now, again, outside of those big three.

Yeah. Yeah. I mean, the, the ones that I can think of, uh, that, that are maybe more at the Figma scale are just looking at, at, at Figma's track record in the public markets and thinking like, I'm not as good of a business as Figma, and I don't expect to be treated any differently.

And when the stock popped, a lot of founders were thinking, "Oh, if I can get that multiple, I should be public today."

I'm thinking like the, the Ripplings and the Deel.

Yeah. All those companies.

Deel's got its own issues, I think.

Which might be separate, but like, but by the way, this isn't just a venture issue though. Private equity firms, which, you know, more mature, slower growth, but more mature companies, they haven't been taking their stuff out either. And, and though, and they don't have anywhere else to go except sell it to other private equity firms. It's just this broader non-IPO issue right now.

Yeah. Um, that,

how much, uh, how much are you spending, uh, are you spending time tracking the, the data center ban that, uh, that Sanders, uh, has sort of proposed?

It's also a bunch of different states. Yeah,

it's a bunch of different states. We had a, we had a big, uh, AI event, um, in DC two weeks ago and I wrote about this a little bit. And on the sidelines, I spoke to the CEO of Constellation, which is one of the big, uh, electricity providers, the data centers, and, and I asked him, this was, you know, whatever, three weeks into the war and I, and oil prices were spiking, and I said, how much, you know, what's this doing to energy deal making in the United States right now, just the, the rise in oil prices? He said it's having a little impact. He said it's the data center ban proposals that are having the really big impact. That's what has people freaked out. I mean, Sanders on the national level, you're, you're not going to get anywhere nationally on this, but on a state-by-state level, you, you don't need much. You need a couple states to do it. It is, it is something that, that opponents have done a really good job on the PR and the industry has done a very bad job on the PR on this.

Yeah.

And, and it doesn't help that everybody's gas prices and home heating oil prices and places where that's relevant and electricity prices are all going up. Iran is obviously exacerbating it. So, but you know, if you're looking at your bill and this is something that is top of mind, well, the bills are only getting, getting higher.

What do you think the impact of the war in Iran will be on defense tech investing?

I wrote about this today. I mean, we have to see what happens, right? It's a pretty fragile peace. It's unclear whether Hormuz is even open or not. Now, it was, maybe it's not. Uh, you know, what I wrote this morning was, I mean, I, I thought a lot yesterday after Trump's civilization tweet, right, that, that if he really went through with what he said, and I know some people are making some odd, um, arguments that, oh, you know, well, it's either a nuke or not a nuke. No, there, there's a lot in between, right? You bomb some major power plants that are for civilians or desalination plants or the power to desalination plants and people don't get water anymore, let alone can't do crops or feed, or feed animals, etc. Um, I, you know, defense tech has boomed in terms of venture capital, which is such, there was almost none of it, you know, seven or eight years ago and there's so much of it now. I think it could have turned Silicon Valley again back against defense tech if the United States had done something that a lot of people viewed as a war crime or, or at least as inhumane. Uh, I think the fact that we have a ceasefire and Trump didn't go through with that, uh, I, I think is probably a bit of a save for defense tech in the midst of a boom. I, I think you've got this, this huge upsurge in all sorts of defense companies that could have actually come to a halt pretty quickly. Not all firms, and still would have invested. Founders Fund still would have, but that, that broader swath of venture capital that fills out those rounds, I think might have slowed down if the military had done something or the Pentagon had done something that, that a lot of people viewed as, as morally indefensible.

Yeah. Something I thought was notable is that the, the American version of the Shahed was a government program and they use some private, you know, contractors for it. But the smartest, the smartest defense tech play

three years ago was just to make the Shahed,

copy the,

and make a lot of them. And it seems like no company actually had the, had the foresight to, to do that. And it ultimately had to be led, uh, from, uh, from the DOD,

I guess. Yeah. Yeah. Interesting. And by the way, I, I haven't looked to see what's happened with the stocks, but I mean, something that is going to have to happen no matter what happens in Iran. Ceasefire, no ceasefire, the, the standard munitions stockpiles are going to have to get refilled, right? We are using a lot of bombs and a, and a lot of, a lot of stuff that's going to have to all get refilled. And I know, you know, Palmer Luckey and all others have talked about how we don't have enough of that prior to all of this happening. And, and that, you know, that, that, you know, that stuff has to get done and there's going to be people who sell that.

Yeah. Yeah, that makes a lot of sense. Uh, well, thank you so much for taking the time to come chat with us. The Axios Show is live and available everywhere, I'm sure. Go check it out and we will talk to you soon, Dan. Have a great rest of your day.

That's the scoop. Great to see you.

Thank you.

Uh,

up next, we have Lior Susan from Eclipse. He's the founder and CEO. We'll be talking to him about, uh, Eclipse's 1.3 billion raise as industrial tech shifts from innovation to scaled production with companies like Cerebras and Forms. How you doing?

Welcome to the show. Uh, please kick us off with an introduction and, uh, and some background.

Let's hit the raise. Good place to start.

Congratulations. Uh, what was the, what was the prehistory of the, of the firm? What, how did you get into investing? What was the first deal? How big was the first fund? Tell me some background.

That's not bad though. How did you set that up? I feel like a lot of people start at 20 or 50. 125 is not bad.

Yeah. Uh, talk to me about the Cerebras investment. How'd you meet the founder? I, it was one of these companies that I'd heard of and, uh, I'd seen a lot of, of negative takes saying that it was the wrong path, that it wouldn't apply to where the current models are going. And then I tried it and it was really fast and it just felt like magical. And so I was all of a sudden converted to, uh, be very excited about the company where before I was sort of uncertain about how it would pencil out. It felt like there was a lot to be done. But you obviously invested early. How did that come together?

Yeah. Yeah. It seems like it's in a fantastic position now. Uh, talk to me about Vulcan Forms. What was the back history there?

Sure.

So yeah, it's a, it's a big fund. You're using, I imagine you take board seats, you lead rounds. Is that roughly correct? Like, where, how many companies do you want in the portfolio? How deeply do you want to be involved? Uh, do you try and pick a single winner in a category or do you, uh, look at more like secular trends and try and get, uh, you know, a broad exposure to the whole category? What's your thesis?

Jordi,

very cool. Um, where [clears throat] do you think robotics is overhyped and where do you think it's underhyped right now?

Yeah. Are doing, doing, I'm assuming there's instances where a company you guys may have backed a company in a category, you know, eight years ago and then a new company gets formed in the category and within, you know, six months they're valued at, at the same, uh, price even though there's wildly different, uh, from a kind of, uh, uh, technical progress standpoint.

Uh, from your position on boards, I, I don't know how much you can talk about this, but I'd love to know your view and expectations for the IPO window. We were just talking about it with Dan Primack. Uh, a lot of, a lot of attention paid to the big three, SpaceX, OpenAI, Anthropic, but what else are you seeing in terms of how companies are gearing up for the IPO window?

Yeah.

Yeah. What are you tracking in terms of trends in the lunar economy? It does feel like we're at a turning point moment, um, with SpaceX and, and Starship coming online. Uh, there's lots of interesting

Oh, yeah. It's not lunar economy. Orbital economy is the buzzword I was

Yeah, any, any, any variation from, uh, low Earth orbit to beyond some of this stuff. When you talk about mass driver on the moon, it starts to seem 10 years away, 20 years away, harder to underwrite, harder to think about, but maybe as a venture capitalist you can start thinking about it. Uh, what, what are you looking for in, uh, in just space broadly?

Well, congratulations on the fundraise. Congratulations on the progress and thank you so much for taking the time to come chat with us. We'll talk to you soon.

Cheers.

Up next, we will be revisiting the Axios npm package hack that happened, a supply chain attack. Uh, npm, the Axios npm was, uh, downloaded 100 million times per week and it was compromised by North Korean threat actors. We talked about it a little bit on the show last week. Revisiting it with, uh, Feross Aboukhadijeh. [music] I hope I pronounced that correctly.

The problem is, is that right?

Uh, yes. So, uh,

We can ask him.

Yeah, we can ask him. Socket detected the malicious update within six minutes and we are lucky to have Feross join us.

What were you guys doing for the six minutes?

Asleep at the wheel. No, I'm kidding.

No, it's, it's fantastic. I'm just

Uh, but yeah, maybe, maybe zoom out and tell us about like the actual process that Socket runs, your business, uh, how the system works and how you're able to detect, uh, supply chain hacks and cybersecurity threats so quickly.

Yeah.

So can you, can you talk about the shape of the threat that was posed by the Axios supply chain attack? Like, because there's a wide range of, you know, zero-day exploit that gives you full access to someone's device or computer or system, uh, all the way to just something that, okay, it would crash if this was, if this exploit was used, right?

Interesting. Then, then in terms of once they get control over, they phish a particular credential, a particular device for a developer who has access to, uh, push changes to a package like Axios. What are they actually changing in Axios to create a vulnerability in the supply chain, in the software supply chain?

Um,

How, what happened with LiteLLM and, and how, like, do, do you have a good sense of how that, uh, contributed to the breach at, at Merrow?

Because, um, the group claims, Team PCP claims that they've, uh, stolen 300 gigabytes of compressed credentials. So that's, you know, that, I mean, think about that, 300 gigabytes of stolen

Yeah. So why, why the boom in the last six months? Is it, it feels like it must be tied to vibe coding or, or AI agents? Um, is this that they have more powerful tools so they're able to do more damage, or is it because our systems are getting weaker because we're pushing more vibe code to production? Is it both? Like, what are, what, what got us to this place where we see this takeoff in, uh, cybersecurity threats?

Yeah.