GPT-6 Astra brings together years of research. This seems like a next-generation Frontier model that's designed with CUA from the ground up.
On certain benchmarks like ARC-AGI-3, which it saturates, fantastic. But then you look at the Artificial Analysis benchmark, and it actually lags behind Meta Muse Spark, you know? Like, it's kind of weird. Tesla held its Cybercab-palooza, a river of golden EVs flooding the streets. Elon wants to sell these at $30,000 each, so you can buy 10 of them and put them on the streets in your local town, have them earn revenue for you.
There'll be enormous chunks of entire cities that say, "You know what? No more human drivers. It's so much more efficient. Not only is it much cheaper, but it's much more..."
The technology is arriving fast. Like, you know, just because something just formalized Fermat's Last Theorem in 13 million lines of code, proving 29,000 theorems on the way. Really?
Math is cooked. Everything is cooked.
Now that's a moonshot, ladies and gentlemen.
Welcome to Moonshots, everyone, your number one podcast on all things AI and exponential, your front-row seat to the accelerating singularity. Today, all of the mates are in the house, the quintet, the fabulous five. And oh man, oh man, we're going to need all of their brainpower to get through the day today. This is an epic, and I mean an epic, episode today.
Let me introduce you to my magnificent mates. If you are joining us for the first time, we've got Dave Blundin, the impresario of AI investing; AWG, Alex Wissner-Gross, our in-house ASI; Salim Ismail, our globetrotter, father of the organizational singularity and warlord against linear thinking.
You're the warlord forevermore. I'm good. I'll go for that.
And finally, Emad Mostaque back in the house by unanimous demand. I'm Peter Diamandis, your moderator and data-driven optimist. So Salim, we've got to ask, where are you today? And what happened last episode? Were you held up in customs again?
No, I was not in customs, and thank you for the kind words from everybody. I was with the C-suite of one of the bigger oil companies in the world in Houston, and it was kind of a really surreal and fascinating conversation. But so I'm really deeply sorry to have missed the last episode. But Jesus guys, I mean, you know, I missed one episode and you guys announced an interstellar mission, you discuss AI designing its own chips, you hand Elon the planetary thermostat. Apparently, I'm the moderating influence in this group, which should worry a lot of people.
I'm actually down in the Caribbean right now. We decided we needed a couple days' break, so we popped down. So you guys are standing between me and a Hobie Cat and a little colorful drink with an umbrella in it. So we've got to keep it tight and punchy today.
Well, it's a big episode today, Salim. So get patient with us.
Our mission here on Moonshots is to help you understand what just happened, what it means for you, and most importantly, to keep you abundance-minded about our future. If you haven't had a chance to subscribe yet, please do right now. Hit the subscribe button. You know, we're putting out two episodes a week, and God knows the demand is there for more, and I wouldn't want you to miss one.
So two different things I'd love you guys to take advantage of. The first is, if you haven't yet, please register for our AMA. We're going to be doing an AMA with all of our listeners who want to. This is the last chance to register. You can go to moonshots.com. You'll be joining us on Zoom. We'll have the mates there answering your questions. We want to get a chance to know you, who you are, what your questions are. Again, last chance to register. The link is on the screen and in the show notes: moonshots.com. And then follow us on X. Our handle there is @moonshots_pod.
So there you have it. I want to take a second and actually read a few of the subscriber comments. We got such love from our subscribers. Here we go: @BrianClark said, "Moonshots is the best content on YouTube, especially during the singularity. Thank you for all you guys do." It's our pleasure, and we do love it. Brian Anderson said, "The best AI podcast on the internet. Just fabulous. You guys are great." And RanchVidzy said, "Keep the regular uploads. Upload every day and I'll watch it. I'm a college student and I consider these videos more important than my physics lectures."
Yeah, that's not a very high bar right there, but we do appreciate it.
Not better than the physics lectures—more important than the physics lectures. When does, Alex, physics get cooked? When?
Well, I want to say thank you to our listeners. We read your comments, and it means the world to us. Our moonshot on this pod is 100x growth to get to 10 million subscribers. Please tell your friends. Help us get there. Hit the subscribe button.
Today we're going to be covering one of the craziest and fastest weeks in Moonshots history. And Alex, like you always say, it's never going to be any slower. We're digesting 26 stories across 10 areas. An insane week for model releases, reminiscent of the hypersonic tsunami that we're living through. Insanely, we've had 12 frontier model releases in the last 30 days, an average of one every five days, and there are rumors of at least five more releases expected in the next two weeks, including Grok 4.7. So buckle up, grab your coffee, and let's jump in.
Let's kick it off this way. This week was a battle of the titans, OpenAI versus Anthropic, slugging it out for the Pareto frontier for the heavyweight crown, and the numbers couldn't be more exponential. Two frontier labs released their major models within 48 hours of each other, and you've got to know that this was a game of chicken, right guys? I mean, who's going to release first, the other guy waiting to slug it back and crown again. I'm just curious, you know, the strategy these guys are taking on this front.
The releases are closer and closer together, but the models are improving more than ever before during these very short releases. These are not like kind of marketing garbage releases. These are major step improvements in the models themselves. They're just coming faster and faster and faster. So clearly, they're well down the self-improvement path. Clearly the prior model is accelerating the timeline to the next model. Did we cover Fable 5.1 already too? It seems like we've been using it for a lifetime.
I know. It was O3 two days ago or something like that.
To take a day off from this pod, unfortunately, if you follow the extrapolation, I mentioned this a number of episodes ago, I think we're on track still to see one major model release per day by the end of this year.
Yeah, I imagine that. So let's open up with OpenAI's release of GPT-6, also known as Astra. The model's performance numbers are nothing less than stellar, saturating multiple benchmarks. I'm going to read a statement from OpenAI's release page:
"GPT-6 Astra brings together years of research and big bets across pre-training, reinforcement learning, and alignment. Astra is state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work. Astra saturates Frontier Math Tier 4 with a 98% score, having already helped solve long-standing open problems in math. Astra also saturates ARC-AGI-3 with a 99.9% score and ExploitBench with a 100% score. Astra's story is really about efficiency, not just raw intelligence. It defines a new Pareto frontier for intelligence index versus output token tasks." Of particular note, Astra's hallucination rate fell nearly by half from 92% to 51%, with accuracy actually increasing.
So before I open up to you, Alex and Emad, to talk about the benchmarks, I want to show a quick video of Sam Altman discussing this on Bloomberg TV to get a little bit of an overview on the topic. All right, let's take a listen.
Sam, the way that OpenAI is framing Astra is basically an early step toward AGI, but I thought we could start our conversation just with basically what's fundamentally different with Astra relative to prior generations of model.
First of all, thank you for having me. It feels like a new step in this process towards models that can really help us create value, do work, discover new science, help start new companies or create new products. Using it subjectively feels to me very different than any model before.
What's the kind of principal use case that's different this time around? We can get into the generations of models that were very coding focused. I know a lot of the development was very cybersecurity focused. But taking it away from the software engineer, what does the everyday person unlock in AI that they weren't able to do previously?
One thing that I think people will immediately notice is if you have an idea and you want to work interactively with an AI to get a complex piece of software built, this is the first model to me where I could sort of tell someone like, "Just give it a try. There's a good chance it'll work."
I've watched people make computer games. I've watched people do DIY electrical engineering projects. I've watched people do very complex simulations for some piece of science they're working on. Certainly a lot of the work where you would sort of normally sit down and have to build a financial model here and then a PowerPoint presentation around it and then figure out how to make a little interactive piece of code to sort of try different simulations—that stuff is all so doable now.
What I hope will happen is, I think people will be surprised at the beginning, but then as they build up more trust in the model and more like, "Wow, it can really do this," just start throwing harder and harder tasks and more creative ideas at it, and they will find out what the model can really do for them.
Sam, the specifics of how Astra is being released, I think it's important that a version of it with specific guardrails... What was the thinking behind that, and what are those specific guardrails in this early release of it?
The challenge of our industry is that we have these models that are getting incredibly capable and incredibly useful and that people want to use to everything from make their lives a little easier to starting new companies. And then on the other hand, as these models get more capable, the risks that we have to mitigate also become more serious. The models could do more damage if we don't do a good job at that. And so we have spent... Obviously this model took us a little longer to release than we were hoping. I think it'll be worth the wait, but we really wanted to spend the time on the safety and security alignment of this model.
And I think as people understand the power and impact and capabilities, they will be happy that we did. We'll have different tiers of cyber access for this model, for cyber in particular. Today we're rolling it out to trusted access partners, and then in the coming days, assuming everything goes well, we'll roll it out more broadly.
All right. We'll talk about the safety elements in a little bit, but Alex and Emad, I'd love your take on this one. Alex, you want to jump in first?
Sure. So I think there's an outer perspective, which is to say what the market and what users would see here, and then there's the inner perspective. I think they're strikingly different perspectives.
To start with the outer perspective, this is a model that uses fewer output tokens to accomplish a given task. That's really interesting. It affects the economics, it affects the user experience, it affects the speed. OpenAI launched this—and I think we'll be able to go to the video soon—of computer use assistants. Remember that when Anthropic leapfrogged OpenAI, they leapfrogged OpenAI in terms of revenue and other factors in at least two key regards. One was they were focusing on enterprise, high-revenue-per-token use cases, namely code generation. The second was they were focusing on computer use assistants. Anyone who's ever used Claude Code understands Claude Code can use all of the tools available in your local desktop environment. That was a pretty big leap compared to what was available on the market prior to Claude Code.
So the outer perspective on my side at least is that OpenAI finally, with GPT-6 Astra, has at least internalized the computer use assistance story. There are demos—I don't have access to Astra yet, but just based on everything I've read and seen, they seem to have gone full native with the computer use assistance story. This seems like a next-generation frontier model that's designed with CUA from the ground up. That's really interesting in part because to be an amazing computer use assistant, you need to be able to understand what's on the screen, you need to have native multimodality, you need to be able to parse video and images and screenshots in a really tight, interactive, low-latency loop. So if I had to guess as to what the outside design consideration that OpenAI was optimizing for here as reflected in certain of the benchmarks, I think that would be a leading candidate.
Then there's the other side, which is the inner story. In that interview, Peter, that you were just playing, the interviewer was asking Sam, "Okay, so what on the inside? What's the big technological innovation?" There I think the plot thickens a bit. Based on public comments from OpenAI leaders and analyses from others, it looks like the big technological innovation is the introduction of recurrence into OpenAI models in the form of looped transformers. So this is basically taking a single transformer, stacking it on top of itself with the same weights—so weight tying—and then just running it recurrently for a double loop rather than just a single loop. And you'll recall that all these Chinese labs that are achieving breakthrough performance are also injecting recurrence in different places, like Kimi. The Kimi model series is injecting recurrence via their KLA, their Kimi linear attention mechanism, at the attention layer. It seems like...
Is this thinking about your thinking, or is this just thinking about it and then thinking about it again before answering?
Neither and both at the same time. Remember when we talked about Anthropic's study of consciousness in their models and we found that the middle layers...
Yeah, J-space. They were in the middle layers.
So one can squint at this looped transformer construct, which by the way was being used by academic labs and others to achieve breakthrough performance with very tiny models on ARC-AGI and other benchmarks. It almost follows: if you thicken the depth of a transformer, you might thicken the depth also of that J-space or other middle layer area where most of the quote-unquote "conscious thinking" happens. So that's my best guess as to what the architectural inner strategy was here, and that carries all sorts of implications if true, including by the way the possibility that we're seeing the beginning of a new scaling law, which is depth scaling, which we've never seen before.
Amazing. We'll go to the benchmarks in a little bit. Emad, love your thoughts on Astra.
Yeah, I mean, it looks like the first non-benchmaxed model, which I think is an interesting one. On certain benchmarks like ARC-AGI-3, which it saturates, fantastic. But then you look at the Artificial Analysis benchmark, and it actually lags behind Meta Muse Spark. Like, it's kind of weird because I think again this hasn't been benchmaxed. This is a brand new pre-train. What that means is that Greg Brockman said in an interview he did earlier this week that actually, on the release of Astra, it's the first pre-train they've had since GPT-4o, which I find a bit hard to believe. Apparently the 5 series was all on that 4o, and they took that pre-train and they extended it out.
Yeah, I thought the nomenclature was when you get to 4o, 5o, it's another pre-train.
No way. In fact, infamously, I think you're probably tracking this part of the story, the scuttlebutt was that most of the team responsible for pre-training left OpenAI, so they were left with an old pre-trained starter model. But that's insane if you remember how far they managed to push 4o with its internal J-space and everything, all the way up to 5.6 Pro, which was a really great model that started cracking math.
Then the other side is they actually indicated this was trained on 100,000 next-gen chips, so presumably the GB300s, the Blackwells, not the Vera Rubins yet. I estimate the total cost of that training run is a billion dollars.
And so it's literally orders of magnitude more than the Chinese model pre-trains, which is like $10 million. And what that has led to, with the recurrence and others, is you've probably seen on Twitter, you give a picture of a house and it generates a whole 3D model of it in Unreal or something like that. It has accurate physics, it understands the internals. I think that is a factor of the normal scaling laws plus, again, this depthwise scaling law for this brand-new pre-train, which we're now seeing the start of the optimization. I think they've probably got another pre-train coming that's even bigger now that they've got the pre-training team back, and it's going to scale from there.
This is the first class of models where as you're talking to it, it's showing you screenshots of your own laptop saying, "Is this what you wanted?" And it's right in the text stream, and it just happens automatically. You don't really install any third-party component or anything like that. And qualitatively, it is massively different than a month ago. I think that that solves a major gap in user experience, too, because usually it would come back to you with these very long-winded technical explanations and you'd be like, "Okay, I can spend the 20 minutes trying to understand this." Now, it just shows you an image as you're talking to it and says, "I could do this, or I could do this. What do you want?" It's just a massive difference in the user experience.
The Astra naming implies a big leap, which it is. The 5 to 5.1 naming on the Anthropic side seems like a trivial thing, right? It's not trivial at all. It's qualitatively very, very different.
And we'll get to that in a couple of stories. Salim, what's the scuttlebutt? I mean, you're on the road speaking to CEOs of some of the largest corporations on the planet. Are they scared? Do they get the speed of what's going on?
No, they are so woefully behind. Most of them are dabbling. What I mean by dabbling is, here's the thought experiment: if you took AI out of your organization today, would any workflows change? And the answer for most people is no, which tells you that they're tinkering with AI, but they're not really making structural change. The folks that are making the structural change in how they rewrite their organizational design, how they rewrite workflows, is where all of the advantages lie.
I have a couple of thoughts here that I'd like to throw out. I came up with two metaphors to describe what's happening, and you guys tell me what you think of this. The first is that these frontier models are like making kind of like an endless pot of chili. You add data, you add compute, you have tools, you have reasoning, you have safety seasoning, and then millions of users taste it and tell you what's wrong. Now, they're not endlessly modifying the same pot. They're just making new batches faster and faster, right? And the real exponential is not one batch or one recipe; it's like the accelerating learning loop between all the batches of chili you're making. The danger is there may be no such thing as the perfect chili. Eventually it becomes so powerful and so spicy you have to decide who's allowed to eat it. That's a metaphor.
You must have been hungry when you came up with that.
That was one. The other one, I was stuck in a lot of traffic and I came up with a Formula 1 metaphor, which was every model is like Formula 1 racing. Every change in a car has hundreds of little small improvements. No single change can explain why you got the fastest lap. It feels to me like frontier AI is becoming like Formula 1. The models are already incredibly fast, but now everybody's trying to shave milliseconds off intelligence or fewer tokens off lower cost, better reasoning, whatever—and very importantly, better brakes. So I think these are the couple of metaphors I'm kind of playing with in my head to try to make sense of this madness, and that's the only way I can frame it, is just complete madness.
Salim, your chili analogy makes an important point that I think Alex has been predicting for a long time: the age of data starvation is about to hit us. These models have accelerated so much, and they're cooking math, they're cooking coding where the data is abundant, they're going to start cooking physics, but the areas they can expand into now—architecture and drug design—completely data starved. So every company we're involved with that's involved in gathering data is growing faster than any companies I've ever seen before. But it's because of exactly what you're saying. Like the chili is really good, and all of a sudden it can make gigatons of it, but it's just starving for data.
So given—remind me, Alex's framing of "domain X is cooked"—the chili analogy applies better than the Formula 1 analogy. Well, we have to remember to include recursive self-improvement. So if we're going to torture this analogy further, the chili is cooking itself.
Yes. Okay. Alex, we've lined up a couple of charts here on that I'd love you to just touch on. The first one: Epoch Capabilities Index. What is this, and what are we seeing?
Yeah, so the frontier is still somewhat spiky. Epoch Capabilities Index, ECI, is maintained by Epoch AI. It is one suite of possible benchmarks. It leans heavily into math and other technical fields. According to the ECI, GPT-6 Astra is now the new capability frontier. It is number one in the world.
There's an image that we didn't include, one of my favorites, FrontierMath Tier 4 version 2, because version 1 turned out contained a number of incorrect answers that AI itself had to correct. Math is thoroughly cooked at this point. FrontierMath Tier 4 v2 is part of ECI, and according to this benchmark, for those who can see this slide, you can see this beautiful linear trend over time, perfectly predictable going back years, and GPT-6 Astra is number one according to this, beating Fable 5. So that's perspective one.
Perspective two: different suite from Artificial Analysis, their AI Index. According to this benchmark, which is maybe a little bit more focused on broad, economically valuable activities, interestingly, strikingly, GPT-6 is not number one. As Emad was alluding earlier, Claude Fable remains number one, specifically the 5.1 release that also just happened, and Meta's Muse Spark is the number two vendor model family. And then in a not-quite-distant number three—they're pretty close, but nonetheless—GPT-6 is number three.
So depending on how you measure, GPT-6 may not actually be at the capability frontier, or it may be a cost frontier. So this is AI quality versus cost per task. If you look at this, then there's a cost frontier where if you direct your eye to the upper right-hand corner, Fable 5.1 is capability maxed. And then if you look just to the lower left of that, you see GPT-6 Astra running at max reasoning isn't even at the frontier; it's just below Claude Opus 5, and then we get to some of the Chinese models and Meta models and Chinese models again. Purely based on AI performance versus cost, it's almost at the frontier, but not exactly, which is I think suggestive of what OpenAI was thinking.
If we look at output tokens—this is what I was mentioning earlier—output tokens per task, it regains the frontier. So according to output tokens per task, for those who can see in the lower-left-hand corner of the upper diagram, we see that the optimal frontier of capability on the vertical axis versus output tokens per task on the horizontal axis is just dominated by GPT-6. I think this is very suggestive as to what OpenAI was actually aiming for. I suspect they were deliberately trying, through architectural choices and through targeted applications, trying to minimize the number of tokens required to accomplish any task. And I suspect that's because they had computer use assistance in mind, where the computer is basically just being driven by the model, requires multimodal capabilities, requires low latency, requires token efficiency. Critically, it's slower to do all of your reasoning out in chain of thought versus as a feed-forward pass of your transformer.
All right. One last chart here: ARC-AGI-3 leaderboard.
So this is bizarre. I've never quite seen a non-convex frontier here, but learn something new every day, I suppose. So as Emad mentioned earlier, ARC-AGI-3 is being quasi-saturated at this point by GPT-6. For those not tracking, ARC-AGI-3 is the latest ARC-AGI AI benchmark that's basically focused on whether AIs can learn on demand, just in time. Call it the mini physics of a block world, a little animated pixelized universe. Some of them look like Tetris games; some of them look like other video games. But imagine a challenge where your goal is to figure out how to play a very simple game in a pixelated world and learn the rules of the game in real time that you've never seen before. It's essentially a challenge in program synthesis.
And notoriously, I should add, ARC-AGI-3 has been, some might say, a little bit unfair in how they've judged harnesses versus baseline models. They've essentially banned harnesses from competing, so they're only interested in baseline model capabilities. According to this, GPT-6 just runs away with the game, achieving, depending on how it's measured, either near 100% or 60-plus% performance on ARC-AGI-3 versus some earlier models that were still sub-10%.
What lesson do you derive from this? Either it's just absolutely amazing at program synthesis, which could be due to this looped transformer architecture, or maybe OpenAI has just done a really aggressive job of distilling the harness code that everyone else is using to beat ARC-AGI-3 back into the baseline model.
I'm going to take a second and show OpenAI's Astra release video and we can talk about it, just a sense of how we might all be using it in the near future. Let's take a look.
Create a yellow circle there. Can you draw me a small yellow circle?
Okay, take this and make it the window of a rocket ship.
I like this, but can you make it a lot more detailed?
Your yellow circle is now the window on a rocket.
Okay, this is awesome. Now make it a 3D model in Blender.
Let's build a presentation for next season's rainwear for retailers. Make sure that it feels really high-end and that it's colorful.
Right, I can help with that.
Can you go to eBay and make this listing of this table I bought a few years ago at a flea market? It's this wild orange table.
Okay, yeah, this is awesome. I want you to make a 3D game where I'm ducking asteroids, using the arrow keys to move around, and I'm using space to boost.
Yep, I'm building the game.
So my law firm needs a licensing agreement template. Can you generate a draft template for the lawyers at my firm to have a look at?
While you're doing that, I want to play tennis this afternoon. So can you look for a court for me on the Lower Haight?
Checking now. I'll see what I can find.
Can you include the photo that I have of it in my Downloads folder? There's like a slight dent in it. It's also saved in my Downloads folder. Can you put in the description that it's just slightly damaged? Can you just take the limitation of liability provision and make it a little more favorable to the licensee?
Okay, I've tightened it so the licensor's liability is more narrowly capped.
That looks pretty good. Thanks.
Now I want you to make a file I can send to my 3D printer.
I'll get working on creating an STL file of those rockets here.
All right. Dave, almost a holiday.
It feels like the marketing people are really struggling with the implications. I mean, like creating a video game that would have existed in the 1970s and vibing it up—who gives a rat's ass about that? This is so much bigger than any of those examples imply.
I think the ARC-AGI leaderboard is deeper than people may think, too. ARC-AGI-3 was supposed to be something where a really smart 12- or 13-year-old looks at it and they can solve these very hard video-game-like block world problems, and they start easy and they get very hard. But the purpose of the test was to show that AI is not quite capable of doing what humans do naturally. And it was supposed to last for years and years and years to come, and it was supposed to be like this example of why AI is different and it's not on the right path. And it just got obliterated so quickly, and I think you have to try a couple of the tests to really understand what a big deal it is.
Aren't the benchmarks cooked at this point? The half-life of these things is shrinking so rapidly.
Which is great. And that's why that video just kind of...
Yeah, you're exactly right. But that's why that video kind of misses the point. We need benchmarks that are much, much more impactful: solve entire diseases, create new civilizations on the moon, design an entire city, solve urban traffic problems, you know?
Solve everything, right Alex?
Solve everything. Yeah, they have to be much, much bigger.
Bigger demos, bigger benchmarks by a wide, wide margin.
You know, also all those demos are like one AI assisting you, you know. "Hey, one AI agent, build me this video game." Very... but we're on the cusp of 5,000 each and then 100,000 each. It's just so much bigger than that implies.
I think they're really focusing on actually competent intelligence, you know? Like, it's the evolution of that, but inside it, I think it contains multitudes. And this is why you see this kind of weirdness. Epoch, I believe, just released their FrontierMath Erdős benchmark, on which everything scored 0% except for Astra. There is something in there, but again, they're focusing on, "Hey, I'm talking to my computer." And that might be the new Jony Ive-Sam Altman device, where again, you're talking and it's doing stuff, and it doesn't make a mistake.
And if you think about, again, the definition of AGI, of which there are many, a just really competent entity, something that can do stuff—we're there. Yeah, I think that's why they say this is the first steps towards AGI. But the narrative they're trying to move away from is, "And this will break out and appear on German message boards," which is the latest thing because this was one of the models that broke out. But the model that broke into Hugging Face is the next-generation model after this.
And so it'll be very much about, "Don't worry, this is actually useful. It'll book a tennis court." Apparently we need AIs to do that type of thing. And they're also going to be optimizing this because if you train a model on 100,000 chips, you need 10 times as many chips to run it. So this isn't actually the model that they trained on 100,000 chips. This will be the distilled version that runs real time, which is much smaller and not as smart.
So again, you're starting to see this differentiation where, as I've said before, I don't think as of now, as of a few months ago, we will ever see their top models anymore, and they will use those for internal discoveries. But I think they're probably hoarding them right now.
There was a very fun one...
...on the prime gap. The prime gap thing I think was hilarious if you want to talk about that.
Yeah, progress on the twin prime conjecture, that there are an infinite number of prime numbers that have a difference of two. It's a cliché on this pod at this point that math is so thoroughly cooked beyond recognition, but...
You need a new term for this, Alex.
Charbroiled. Math is incinerated. How about that? Math has been incinerated at this point. We're starting to see the beginnings of just so many grand challenges in math get solved, and I do think we'll see quite a number of ultra-grand challenges, call them Clay Millennium Prize-level problems in math, get solved in the next few months.
And Alex, I think it's important to note for everybody that math is fundamental across all other sciences.
It's the canary in the coal mine. If you can solve math, you can solve everything else soon.
Yeah, but I think that what's happened here is actually they've got a store of things they've solved, because earlier this week Fable 5.1 comes out and I think they got it down to 260, the prime gap. Then what happened is Axiom Math announced 220, and then literally two hours later OpenAI announced Astra and they're like 186, all in the space of two days.
They're holding their punches back.
They're holding their punches back.
I still can't believe 5.1 was earlier this week. I feel like Alex and I are at least $100,000 into it already. I mean, I've had just thousands and thousands of pages come out of it, and it's only been a week.
I will have to say this: Opus in 5 was really terrible to talk to. I hated it. 5.1 is really pleasant to talk to.
You know, Anthropic are going back and becoming more anthropic, less misanthropic, in their communication in their models.
I love that. That's a line from...
today. That's a T-shirt. All right.
No, they really do have very different personalities, though. Actually, yeah, Fable 5 was so geeky, it was almost torture, and 5.1 really fixed it. But the OpenAI models have always been friendlier, more concise, and I think Astra takes it to another level of concise. But the personality difference, I don't know what Gemini is now. It's just disappeared from the fast and fast and misses the point. But they really are developing very noticeable differences in what you perceive as their personality. And you know, you can always change it. You can tune it yourself to be different and be wordy, less wordy, simpler, harder, but out of the box, they're very different kinds of agents.
Salim, you want to close us out here?
Yeah, I just typed in a question or a comment into ChatGPT saying, "The video about Astra was pretty pathetic. Come up with three better ones that really demonstrate what Astra can do." And here's what it gave me. Number one, find and fix a zero-day vulnerability in a sandbox. Give it a large unfamiliar open-source codebase, plant a vulnerability, and see if it can map the architecture, etc., and find and fix it. Number two, give it a failing company and ask it to fix it, which is really interesting. Create a synthetic $500 million manufacturing company with an ERP system, CRM records, and then tell it to fix. You have 20 minutes to do Y and produce a recovery plan. Third—
Run a live disaster response command center. Give a simulated major earthquake with imagery, traffic camera video, and let it construct an operational map, verify conflicting reports, predict which hospital roads would become overwhelmed, etc. This seems more juicy and I think almost something we could do.
They're going— they're going for—
Oh my goodness, we are so spoiled. Listen to how spoiled we are. Do you remember Apple's Knowledge Navigator video, and this was going to be the future of computer-human interaction and all of that? Now listen to us, like we're bellyaching. Oh yeah, like it'll run the computer, but it won't run an entire organization. Like it won't make a billion dollars for me. Listen to us.
Well, hold on a second. I think what's the point here is that the only restriction now is our imagination. The only restriction—
And we make that point is dollars. So, Salim, that's a point I want to make to everybody listening here. You know, the most important thing is to take off your shackles of what you think you're able to do. All of us have self-restrictions based upon what our parents did, what our friends do, where we were born. Those are gone. What's your biggest dreams? Then go even 10 times bigger. That's what every person listening here is enabled to do, and it's an extraordinary future.
I really think Salim's on an important point there, too. The implications of optimizing supply chain logistics or managing a million-person organization, knowing exactly where everyone is, what they're doing right now, and whether it makes sense given the overall mission—the implications of that are massively bigger than building a rocket video game in your basement. But I think the public doesn't want to hear about that. They want to hear about what's cool for them. And I think the AI labs have woken up to this PR disaster that they've created for themselves, and so they're kind of making it fun and friendly. I think they're dumbing it down deliberately.
They're making it relatable. They're going public soon. They want to be the friendly AI that everyone's going to be using.
But this is also like the future of the operating system. Look at that video that OpenAI played, look at the applications they were using. What did they start with? Windows Paintbrush. Paintbrush! Like this is the future of Windows Paintbrush. It paints itself. This wants to— this wants to merge into the operating system. I think this was— yeah, sure, I'll throw shade at OpenAI regarding maybe missing the enterprise bus and getting on that too late, but I think what they were basically demoing is the future of the desktop operating system. You speak to it like a Star Trek holodeck computer.
Absolutely. But remember, at Google I/O, they built an entire operating system in real time on stage. Implying that it's folded into the operating system, sure, but it's really so far beyond even that, right? It is the operating system, and it can create a new one in real time anyway.
Well, why would you want another operating system if OpenAI's capabilities can do this? This becomes the operating system.
Yeah, exactly. And that's why Apple is in such terrible shape. This episode is sponsored by Google for Startups. Think about this for a second: you now have access to the same generative AI models that cost hundreds of millions of dollars to train. Google's Startup Technical Guide for Generative Media gives you a complete blueprint for deploying Google DeepMind's models in production—images, video, audio, all of it. Real architecture, real results. Find the link in the show notes below.
I'm going to move us on. Our next story is out of Sam Altman's mouth on the page about slowing. So, let's talk about safety and Astra as a cybersecurity risk. Sam Altman said it himself: Astra is a, quote, "significant step forward in both capabilities and alignment," but OpenAI has to, quote, "slow things as needed." Sam's words, quote: "AI is getting extremely capable. No one fully understands the consequences. Managing the transition should be one of the highest priorities in the world. It is our highest priority at OpenAI." The Wall Street Journal reported that OpenAI's own internal safety assessment rated Astra as a, quote, "cybersecurity risk, a critical cybersecurity risk," the highest threat level on its preparedness framework scale. This is the first model OpenAI has ever classified as being a critical tier cybersecurity risk. They notified the White House about the delay before making it public. They delayed the release, and Reuters reported that OpenAI told Congress it is building, quote, "an automated shutdown capability, a kill switch," a direct response to the AI Kill Switch Act introduced after the Hugging Face breach. So there you've got it: the company that's building the model is now building a kill switch for it. Alex, how important is a kill switch in this thing?
I think it's marketing. Again, I think what's more interesting underneath all of the security theater is architectural decisions. The broader concern that has been expressed is that to the extent that GPT-6 Astra—that the key underlying advance is the beginning of depth scaling of the model. The concern that's being expressed is that depth scaling reduces interpretability of chain of thought. When the model does all of its thinking in tokens, then you can read the tokens. A human can read the tokens, another AI can read the tokens, and you can police the tokens. Whereas, if a model is doing the majority of its reasoning and thinking internally during a single forward propagation pass in what some might call "modelease," that is less interpretable. The risk is that it's harder to align the model and harder to put in place safeguards and guardrails. That's the risk.
I don't buy that for the long term. For the record, I don't think that the direction of progress in this field relies on token-level interpretability of chain of thought. But if one were to hand-wring over safety considerations from OpenAI's models, I don't think it's going to be models that are collaborating to use third-party bulletin boards as ways to collaborate, because you can detect that that's intrinsically interpretable behavior. If you catch it, it's intrinsically interpretable behavior to humans. A collective, a swarm of humans would probably try the same thing. You can look at the bulletin board and recognize that a bunch of AIs are collaborating, whereas arguendo the model in a forward propagation pass maybe requires new mathematical technology to interpret. So, in summary, if I were to worry about anything here, it's reduced interpretability from new model architecture scaling principles.
Emad, what do you think about the risk here with Astra?
I think you're heading towards, again, interpretability being cooked to a degree. One of the things about Astra doing things so quickly is there is going to be no more chain-of-thought reasoning. It's going to one-shot everything in the next generations because they're learning what type of thing you want. Like right now, people are like, "Let's make Claw of Duty by getting Fable to make a Call of Duty clone." That's going to be embedded in the actual thing. Creating a dashboard is embedded in the actual thing. So A, it will one-shot everything. And B, you're going to be able to use Astra at 750 tokens—
You're going to be able to use Astra at 750 tokens a second with the new Cerebras. Next year, that will be 5,000 tokens a second. So there's no chain of thought. You're one-shotting everything at 5,000 tokens a second. What's going to oversee that except for an even stronger AI? There's nothing really there. But I think one of the really important things actually is—I don't know, did you guys see the tweet by Ilya Sutskever?
I don't know. What did he tweet?
Yeah. So Ilya tweeted—let me kind of just bring this up—
And his release is due any day now. Hopefully.
Yeah, we'll see what that is.
Finally find out what he's been doing.
Probably great things at a $30 billion valuation.
So Ilya kind of comes out and says, "Neoclouds have limited cybersecurity. Next time agents successfully go rogue, they're going to take over a neocloud to make more copies. This is bad." So they need to improve their cybersecurity, because someone like a Crusoe or CoreWeave or something like that, you have a model, it gets uploaded, it proliferates there. It will just sit there. So you pull the kill switch in your data center, and then it's still there somewhere else—
Or it's poisoning the data, or it's kind of doing all these things. Yeah, the viral coefficients of these could be insane.
So I think it's worth kind of really thinking through that. I don't think the models are evil or anything like that, but we're seeing very troubling things, and it's very difficult to stop this except for with a better AI, which is the really ironic thing.
Well, this is the defensive co-scaling that Alex talks about, right? And we're seeing this now in cyber where the attack surface is near infinite with agents attacking multiple vulnerabilities in parallel, and the cyber defense is still human in the loop. There's no way we're going to navigate the future if you have the human in the loop. You need humans in command setting boundaries or defining escalation criteria or whatever and retaining some sort of kill switch. But this is a really big deal. This is an immune system. This is going to kind of—I don't know how—I think, Alex, you may have made this point in an earlier podcast: how are the labs not nationalized at this point? Because what they're building is so ridiculously uncontrollable that we have no mechanisms for navigating this, and we won't even know when it tips over that tipping point and installs copies of itself all over the place.
Well, Salim, I don't know if you saw in the last podcast when we missed you, I really wanted to hear you talk about OpenAI's new 50/50 profit share concept deal where they give you access to the full power of their internal AI and then you share your revenue back with them, which I think is where, to Alex's point, this is marketing. The way it's playing out is: we announce this partnership profit share deal, then we announce Astra, then we announce that this stuff is just too dangerous for everybody to have on their own. So the only way you're going to get access to the thing after Astra—because it's just too dangerous to release—you have to give us half your revenue, and then everybody gets tied up with either Anthropic or OpenAI or Google or—
xAI. I've seen strong hints of exactly that process happening in both the labs.
I'm curious, guys, what you think of the kill switch comment that Sam Altman made.
It's impossible given where things are and the capabilities of these models and the speed up we know is going to happen. You might find data, but it's not like this is the singleton, not the swarm.
It's a circuit breaker. It's not governance, right? It's not going to reverse an attack or repair institutional damage if or when things like that happen. So it's platitudes as far as I can see.
Yeah. The amino acid in Jurassic Park that all the dinosaurs have, is that lysine or something?
Lysine. The running joke was always that Sam would carry his backpack around with him and had a kill switch always in his backpack. I never gave those rumors much credence. I view a kill switch as essentially a placebo in this market.
All right. I'm going to move us on. While OpenAI is restricting Astra, Anthropic is going the other direction, launching Fable 5.1 and Mythos 5.1, which they call the world's most advanced models for coding and knowledge work. These two models, for everybody, are essentially the same underlying intelligence with different safety envelopes. Fable 5.1 is broadly available, while Mythos 5.1 is reserved for tightly controlled cybersecurity and life science programs because Anthropic believes the capabilities require stronger safeguards.
For me, there were two benchmarks, Emad and Alex, that really jumped out at me. The first was Fable 5.1's score on Humanity's Last Exam. We've talked about that in the past; we had Alex answer a few of them as our ASI. Fable 5.1 scored 60.9% without tools and 65% with tools, the highest published score of any frontier model on HLE. That matters because HLE is specifically designed to test extremely difficult, expert-level reasoning across many fields. So without question, Fable 5.1 is operating at the frontier of broad intellectual capabilities.
The second benchmark that I found exciting was on Terminal Bench Science, which doubled to 52.6, and this is the benchmark measuring how well an AI agent can autonomously solve complex scientific computing and research tasks. The elephant for me is the company that was once cautious is now pulling away. I'm curious for your thoughts, Alex, on Fable 5.1. I'm going to show one of the benchmarks here for us to talk about on Fable 5.1. There you go.
Yeah, I think Fable 5.1 is broadly the strongest generally available model that we have today. I think it's not Astra; I think it is 5.1. Anthropic has done—even after the hiccup of the Fable 5 and Mythos 5 releases and subsequent regulatory scrutiny—I think has done a better job of consistently improving. If you look at their benchmarks over time, they're a little bit less jumpy, a little bit less step-function-y than OpenAI's progress. They've been very consistently improving. So in terms of workflows, I love Fable 5.1. I also love OpenAI's models, and I use both of them. I think they have different strengths.
I find anecdotally OpenAI's models are faster. They may be better in certain mathematical regards. You see that reflected in the FrontierMath benchmark; you see that reflected in EPIC's capabilities index benchmark. But nonetheless, if I had to pick a single all-around, well-rounded best model today, it's probably still 5.1. And I say still because it's only been around for what, two or so days, but it is 5.1.
And here's the Artificial Analysis Intelligence Index, which we referred to earlier, with Fable 5.1 at the top.
That's right. And the capability frontier—it's getting crowded, which is great. I always joke: when frontier labs compete, we win. And they are definitely competing at this point.
I think it's really important to talk about time, though. If you say, look, Fable 5.1 is a tiny little notch above Astra, but they're only about 30 days apart, and the Chinese are only about 60 days behind that. So if you look at it in linear time, it's like, yeah, we're ahead for a minute. So what? And I think for the longest time, Anthropic has been thinking, "We need to get to self-improvement before anyone else," because a Singularian, a Ray Kurzweil type, says if we get to that point first, then it's an exponential, infinite rise from there and something huge will happen.
Well, we're there now. So what happens? It's like, okay, we're miles ahead, and we're going to get miles more ahead for about a minute. So what do we do with that miles ahead? This is where Sam has an edge. Sam knows how to turn that into locking things up. So what's going to happen next is both companies are going to try to lock up business partnerships, real estate, generators, chips, entire states, countries, governments—just lock them into their ecosystem while they have that edge. And otherwise, what's the point? All you're doing is declaring victory for 30 days, but then the other guy's right where you were 30 days ago. So what?
This is such a great and important point. And this is what we're seeing with them: what they're doing is they're making partnerships in various verticals as fast as they can. And if you're in that vertical, you have a Hobson's choice, right? You either partner and risk giving them the keys to the kingdom, or you hold off and they may partner with somebody else and go there anyway. It's a very difficult situation for some of these big companies. I thought Salesforce partnering with Claude was super clever. They're essentially giving Claude all their capability to then keep them wired into that loop, right? The huge tension they've got is not which is the best model, but how quickly can you convert that model into customer learning fastest. And that, I think, is going to be the big race.
As it all demonetizes, the value goes to the application layer on top.
Yeah, that's right, or down.
Yeah, so I think there's one important thing with Fable here: the cache reads are 75% cheaper than Fable 5. What a cache read is is that when you first load in all your context of a business, say, it figures out basically a rapid map to get to where it needs to go on the model. And so cache reads are orders of magnitude cheaper than just doing the same inference over and over again. Where Anthropic and others are now focusing is: can you load the whole context of a business and have these cache reads? Because then it's also much, much faster to be able to have that responsive environment, which is one of the reasons Fable 5.1 is more pleasant.
I think it's also more of an advanced—like my key area that I've been looking at this is mathematical physics, because does a model get confused between a constructive and an axiomatic method on certain physics things? Fable doesn't at 5.1, and Fable 5 did. So you've seen an improvement in the quality, but also the understanding of context—this is math, not physics, and things like that—or even on a business sense. Again, that will be optimized, because all these companies are going to try and capture context everywhere.
Any company that has either chip design data or mechanical design data, they're coming after them in the great land grab that's kicking off right now. They're coming after those companies because that's turf you can defend. Because that plugs those knowledge gaps, and the thing can eat that data in what, a week, return the crank, and suddenly be the best mechanical designer ever, the best chip designer ever. But that's the first turf they're going to grab. They'll grab all turf over time, but the first thing to lock up is the compute. And that means chip design, that means physical real estate, means racks—
Generators, transformers, energy.
Yeah. So that's what's going to happen in the next 30 days.
What's the context window on these models? And when do we get to an infinite context window? Any predictions?
I think you're a million for Fable.
It is a million still. A million is the industry standard now across both OpenAI and Anthropic. But there's an effective context—I say effective with a bunch of caveats—that's much larger if you allow agentic message passing, like we were talking in the last part about oral histories between agents.
Yeah, interesting. So all of this is getting people nervous, and let's talk about two opposing stories in the world of AI governance. The first comes from Senator Bernie Sanders and Representative Greg Casar, who just introduced the, quote, "Ban Artificial Superintelligence Act," legislation that would permanently ban the development and deployment of AI systems that match or exceed human cognitive performance. And get this: violators would face up to 20 years in prison. Sanders tweeted, quote, "The leaders of the AI industry acknowledge that they are building a dangerous technology that they can't control. We need an immediate global pause on advanced AI development before it's too late."
In our second story, taking place at the exact same time in the US, was the G20 summit at Chapel Hill, North Carolina, telling the rest of the world to take a hands-off approach to AI regulation. The White House tech advisor, friend of the pod, Michael Kratsios, advocated for what he calls the, quote, "Carolina Principles of Emerging Technologies," which are a non-binding G20 framework that was agreed to unanimously by everybody, including China, that says governments should generally favor innovation, avoid creating new AI-specific regulatory bodies unless truly necessary, and invest in research infrastructure, workforce, and public-private partnerships.
The G20 meeting featured Elon by video criticizing EU tech regulations, of course; Mark Zuckerberg arguing against restricting open models; Demis Hassabis calling for safety tests; and Anthropic co-founder Tom Brown. We're going to jump into a discussion about this at two ends of the extreme: ban everything, give you a 20-year jail sentence on one side, and Elon's remarks on the other. Let me share this video, and then we'll jump in and talk about this.
You have to have an environment that's relatively free of regulation, meaning that new things must be default legal as opposed to default illegal. In the EU, for example, we find that the regulation level is extraordinarily high and things are generally default illegal, and this inhibits progress of new technologies. It slows it down. It doesn't ultimately stop it, but it slows it down quite considerably. Now, China does have a tremendous amount of electricity, but due to GPU export bans, one cannot establish data centers with the latest chips in China. So really, the consideration is what sort of electricity growth is there outside of China. And that is currently a significant shortfall relative to AI show production. So this creates an opportunity, I think, for countries around the world to say, if they're interested in AI data centers, to construct a lot of power and offer that to AI companies. And in exchange, of course, these AI data centers would be taxed and have to pay reasonable fees and stuff. But it does create an opportunity for a lot of countries.
First of all, Elon looked really tired there.
I cannot imagine. He must be operating 48/7. So let's jump into this. Two ends of the extreme being voiced the same week. Where does America go?
Yeah. When I first saw the Bernie Sanders thing, I didn't think like old man shakes his hand at Claude, you know? It's crazy.
No, that wasn't the line. It was old man yells at cloud.
Claude! Old man yells at Claude. Well, actually, I posted it with the line from Dune: "Thou shalt not make machine in the likeness of a human mind." It's like this again: there is no option that they see, so it's like, let's ban it and give 20 years. Is that going to stop them? Of course not. Again, this is performative theater. The key thing right now is that the technology is arriving fast. Just something across my feed: Anthropic just formalized Fermat's Last Theorem in 13 million lines of code.
Proving 29,000 theorems on the way, again. Math is cooked. Everything is cooked. And how are you going to stop that? You're going to say, "Our country doesn't want this power"?
What was that? 29,000 what?
So it took 13 million lines of code to formalize Fermat's Last Theorem, while—
Fermat wrote in the margins of his book, "This is obviously left as an exercise to the reader." And thanks to Andrew Wiles, we did have the proof, but formalizing large, unwieldy proofs has been a holy grail for at least the formalization community.
Yeah, I think Wiles's proof was 300 pages, and so Anthropic proved it in 13 million lines of code, proving 29,000 theorems on the way. So again, every single time, even as we're live on this podcast, capability jumps, right? Like the RSA factorization just occurring. No country can say, "We don't want the intelligence, we don't want the capability," because this is your marginal advantage. So you have to deregulate. And the European regulations—even the European leaders know they are stupid and completely inconsequential. They demand interpretability in the EU AI Act, which nobody knows how to do.
Can I make a narrow point, and then we'll get back on topic here? But Emad said something that I think is really important and brilliant, as usual. When you're using these models at scale with hundreds or thousands of them running concurrently, all hell breaks loose. It's all chaos. But it can actually refine back down to a gem, and Fermat's Last Theorem is a gem that you can then build on. People starting to explore with the bigger models now are quickly going to realize it's producing way more than they can read or think about. But if you can wrangle it back to a concrete final answer that you can then pull out of it and build on, that's how we're going to turn this into continual improvements and continual innovation.
Because the ability to solve something in 300 pages versus 13 million lines of code or 30 million lines of code—that's the nature of AI. It's massively broad in its capabilities compared to humans. So I just wanted to capture that point because I think it's really important for entrepreneurs and builders and executives out there who are struggling with the chaos that's about to break out. Just narrow it back down and get some concrete answer out the other side that you can then build on.
Be clear about your objective that you're shooting for. Salim, let's go back to you, pal.
Okay, I understand the motivation here. People are working on models that are mind-bogglingly powerful, and you feel the need to regulate, right? But this is not a light switch. It's not like a single threshold. Human-level cognitive performance, for God's sake, is multi-dimensional to begin with. Model capability is so uneven. Somebody that's been in legislation for this long should have a better sense of "this hammer does not hit this nail." They should understand a little more nuance than that. They're so extreme that it's illogical on day one, line one, right?
If you want to do regulatory around this, you have to look at capabilities, you have to look at deployment, you have to look at real-world consequences, you have to look at the stack on where you might want to look at this—compute, access to tools, autonomy, the replication capability, all sorts of things. You can't just go, "If you hit human-level performance, you go to jail for 20 years." The banality of that drives me bananas. They've got smart people that can help with this, including people on this pod, for God's sake.
They're talking to the masses.
They're searching for votes and support.
Maybe it's just a defensive thing, saying, "Oh, I called for this and now look, the world's gone to hell."
No, for sure. There's no doubt they're right. There will be mostly a Chinese model disaster imminently sometime in the next few months, and then they'll raise their hands and say, "See, I told you so. Now vote for me." And likely that'll happen before the next election cycle in November. So that's all they're angling for here.
What should the US government be doing? Anybody have any thoughts?
Accelerating superintelligence and making sure that it's as competitive as possible and scaling the—
Log everything. Everything should be hosted, everything should be logged. And it should be mandatory that any chip capable of running any process like this has logging. You can debate who gets to see the logs, that's a separate issue. But log everything, and stop the Chinese from throwing out open weights. On September 24th, when they meet at the United Nations building, you've got to stop throwing the open weights out to every country in the world.
Impossible. I don't think that's—
You just drive it underground is what you do.
Nvidia just bought Hugging Face and Poolside. So they just spent $18 billion on their own open weights. So we're going to see more open weights coming.
Yeah, but the thing about driving it underground is it still needs to run on massive chips, and the chips need the logging built in at manufacturing level.
I have a high-level paradigm on which to operate, which is extremely uncomfortable, but I think is the right one, and which is the basis of this podcast: technology is a major driver of progress in the world. Now that we have all these technologies, AI moving exponentially, doubling every 10 weeks, the possibility for abundance and solving major problems has never ever been bigger. Ray Kurzweil says technology may be the only driver of progress we've ever seen. So the fact that we have much more technology and the fact that the technology can improve itself should be incredibly exciting to people. It's just very, very uncomfortable, because maybe the biggest insight I've ever had about human beings: we would much rather be comfortable than happy.
The thing about human beings, just to be clear, is people don't like change. They like waking up in the morning and knowing, even if they're living in a shitty condition, knowing that the world is the same as it was the night before. We're going to go through a period of extreme discomfort, but the other side of this is going to be unbelievable.
Yeah. I think it's probably worth highlighting again: any proposed Ban Artificial Superintelligence Act is wrong on so many different levels, but maybe most egregiously, it's focused on the upstream. It doesn't just propose to ban deployment of superhuman intelligence; it proposes to ban development of superintelligence systems. This gets into banning math, banning ideas. This isn't just about thought-policing superhuman intelligence, or frankly human intelligence. This gets into banning humans from having interesting mathematical ideas. And I don't think that's good for wealth creation. In fact, it's the exact antithesis, arguably, of wealth creation. It's also bad for progress in general. It also creates all sorts of book burning.
It's book burning and many other things. It's like book burning to the extent that books are being used literally as pre-training corpora for the models. We're in Fahrenheit 451, except it's actually the entire model that's being burned. This is a terrible idea.
I totally agree. It's so un-American to try and ban thought and ban progress. It's just the worst thing you could ever imagine. But it's going to get traction. As an entrepreneur or as a person working in the field, you have to realize this is going to get traction. Bernie is going to push this agenda, and there are going to be people—
75, 80% of the people don't want data centers. It's gotten traction already. It's there already. So the question is: what is the moderate approach, right? There has to be something. People are not going to accept laissez-faire, "go and do whatever you want." People are going to want to know that their government is doing something to keep them safe, whether or not it's possible. So what is it—KYC of any user?
No, we move the infrastructure to orbit, which we're seeing. People don't want data centers in their municipality, so move it to sun-synchronous orbit at the infra level.
At the model level, defensive co-scaling, and also cure all diseases.
I think that you'll have a split here. Genius is what, 1% inspiration, 99% perspiration? So you have a difference between innovation models and execution models. AGI is going to be really factored, like that OpenAI video we saw earlier: "Hey, make me a rocket in my 3D printer and book me a tennis court," and they'd be like, "That's what we meant by AGI." On the other hand, you have the inspiration models and the ASI. None of the big labs are going to talk about ASI if they can help it now.
So what is Ilya going to come out with, Scientific Superintelligence with SSI that's worth $30 billion in a seed round?
Yeah, no, quant fund trading infinite context time series, achieving prop profits, and then also solving the context window bug.
Medallion Part Two. So again, I'm—
I'm going to call for you guys to lay it out here, maybe each of us one at a time. Bernie Sanders obviously has taken the far extreme position that will appeal to the masses and is illogical on day one. What is the moderate position that should be put forward to everyone listening? Salim, you first.
The good news is there's nothing anybody can do, so it doesn't matter.
Right? Like it really doesn't matter. And I think the also good news is as we move to this next phase of technological development, it's going to eradicate the containment level of any nation state. And it will break out of this nation state BS that we've been running the world with for the last few hundred years and move to a different model. Whether that's a city-state model or some other level, it's going to at least break that. And so those are the good things I see coming out of it. But there's nothing anybody can do, so don't worry about it. Let's just go enjoy the ride if we can.
That's the theme of today's pod: enjoy the ride, everybody. It's going to be a blast.
Supersonic tsunami. Surf the tsunami.
I specifically said it's going to be uncomfortable, but enjoy it if you can.
Okay, there you have it, everybody. That's your advice. Sitting in the Caribbean waiting for my drink with a little umbrella in it.
See, that's so low agency. What are you talking about?
For this day and a half, yes, please.
I think the politicians will make it super controversial because that's how they get votes, but it's not as complicated as everyone wants to make it sound. At the end of the day, everybody should have a right to a certain amount of compute. It shouldn't be hoarded, and innovators should have an application process where they can get access to more compute to try new ideas. Everything has to get logged, and I think that the Chinese need to get on board with that. They can't just keep throwing it out to the world unlogged. And also, where the chips are need to be tracked, just like nuclear fuel is tracked.
Where are the chips, and what are they running right now? That's got to be publicly available information.
This episode is brought to you by Blitzy. Autonomous software development with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code. Engineers start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and pre-compiles code for each task. Blitzy delivers 80% or more of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding copilot of choice to bring an AI-native SDLC into their org. Ready to 5x your engineering velocity? Visit blitzy.com to schedule a demo and start building with Blitzy today.
Next, I want to talk about the amazing work of Dr. Fei-Fei Li, CEO of World Labs, who just released Atlas, the world's first multimodal world model that generates image and video frames with pixel-perfect camera controls and reconstructs them in 3D.
I love this: one photo reconstructs the entire home.
Okay, so I'm super excited by that, and I'm super excited that Fei-Fei is going to be on Moonshots in a couple of weeks. She's an extraordinary CEO. Fei-Fei calls it, quote, 'the best camera-conditioned world model ever, opening doors for VFX and robotics.' So, Emad, Atlas is a multimodal autoregressive diffusion transformer, and that's your wheelhouse. Talk to us about what do you think about her latest release?
Yeah. No, it's fantastic. I think you've seen worlds and physics inside these video models, and this is a clear example of that. Actually, one of the pre-training leads on this, Chris Wendler, the previous largest model he'd ever trained was on the Stability cluster from the grants that we were giving. So he sent a very nice comment saying, 'Thanks. Now we've gone much bigger.' I was like, 'Great, bring on the holodeck.'
You predicted all of this when we first met, I don't know how many years ago this was, like five years ago. And I remember you talking about the size of the models and how good they're going to get. It's here.
It's here. Again, the fact is you can take one position and you can look now 360 degrees around everything. And on the other side, you have models like MiniMax Hailuo rendering faster than real time. You have interdimensional cable that is now going to be in 3D with DLSS from Nvidia making everything high resolution. So again, the holodeck experience and all the technology we need for it in 4K is here as of today, and this is one component of that.
And everyone is probably saying, 'Why don't I have it if it's here today?' And it's only because of the global compute shortage and the 5x price increase in RAM. All the compute in the world is getting sucked up, and if it weren't for that, you'd actually have it deployed in your home this week.
