OpenAI Pauses Frontier Training, Elon's 100X Prediction Lands, Robot Beats Usain Bolt with Emad Mostaque | EP#282

OpenAI announced it is voluntarily pausing some of the frontier reinforcement learning training that it's doing. What have they paused, and is it really significant? And do you think the other frontier labs are going to do the same thing?

They're so powerful, even we can't trust them, so we have to throttle back. It's marketing.

Elon Musk's January 6th Moonshots podcast prediction of 100x gains was at the edge of plausibility when he made it. Now it's simply a fact.

Imagine I gave you 10,000 employees tonight. Oh my God, if I had that, I do something amazing. Okay, what? Start thinking about it because it's coming imminently, and it's actually not an easy problem to figure out how to turn it toward creating good.

Unitree's newest humanoid robot broke every human standing jump and speed record. A top speed of 12.66 meters per second, beating the human record set by Usain Bolt.

You don't want to have superhuman robots on the street because you'll have accidents, you'll have issues just like cars. I think that these types of robots will be banned.

Now that's a moonshot, ladies and gentlemen.

Welcome to Moonshots, everyone, your number one podcast on all things AI and exponentials. The news that matters, the news that's changing your life, your front row seat to the accelerating singularity. I'm here once again with my magnificent Moonshot Quintet. Yes, they're all— 5 of us are back. AWG, Dave Blunden, Salim, and Emad. I'm Peter Diamandis, your host, your abundance whisperer, and we've got a lot this week. Gentlemen, good to see you all.

Good morning. Howdy.

Good morning.

Good afternoon in London. It's like everybody's in their normal haunt except for me. I'm up in a sleepy town in the Pacific Northwest trying desperately to take a little bit of time off to think.

Taking shelter from the singularity.

Yeah, except I got up at 5. 5 AM this morning to be with you guys, so what the heck. But I hate the old saying, you'll sleep when you're dead, because I just don't want to die and sleep is still so important. But hey, what can I tell you?

We'll solve sleep soon, don't worry.

Yeah, no, I— and death. I think like death is counterindicated at this point.

Uh, you know, as always everybody, our mission is to help you understand what just happened and what it means for you. And most important, to keep you optimistic about the future. If you're new to this pod, or if you've been a regular, great to have you back. Please take a moment and hit the subscribe button. You know, we publish Moonshots twice per week at a pretty regular cadence. It's great to have Imad here, hopefully at least once a week. And sometimes when there's extraordinary breaking news, we publish 3 times a week. We read your comments and we love you too. I mean, it's been an incredible outpouring of support. I don't know, you guys see it. You know, I get stopped On the street, in the grocery store, people are saying, you know, I live for your show. I love the show. I can't go a week without listening to it. Are you guys getting the same response?

Yeah.

You know what I'm getting a lot of is what we podcast out is so different from other podcasts. But if you go back to shows from 6 months ago and a year ago, people are like, what the hell? They were saying that back then and now it's here. And they don't get that on any other channel. So they're, they're really appreciating the ability to plan around what we're saying. So we got to be accurate, guys.

How about you?

No pressure.

Can I be Eeyore for a second? Sure. I am getting acceleration fatigue. Like, I mean, Jesus, can, can we pause for a week, right? Every morning it's like, our model is 100x better, a robot is running faster than Usain Bolt, like AI is designing proteins, like drones are doing a million deliveries a day. It's like, you know, it And it's tough because we've spent our like half our careers, Peter, you know, I talk about exponential technologies. We love this stuff, but I'm tired. Like, it's we've gone from wow, look what happened like this year to what happened since Tuesday. I know.

So Liam, you looked all the way around. We're supposed to be accelerationists here. You're tired already?

Like, we're we're only part of the way through the singularity. Like, this is maybe just the first inning. I'm really exhausted.

Isn't this slow? It's really easy. It's, it's, uh, you know, this goes back to Peter, right? Our brains evolved for a point, a time when our— like, the world didn't change lifetime to lifetime. Next Tuesday will be totally different. And maybe the hard part for, for us is, is staying human while all of this bubbles up around us.

Oh, so true.

It's not keeping up with the technology, it's like just staying like, like— so anyway, just, uh, Yeah.

What can we do, Salim, to help you with your acceleration fatigue?

It's probably not helping that I'm on a plane every second day. That's probably not helping.

But yeah, but this is the slowest it will ever be. And I know, and, and it's just, you know, the only way I keep up with what's going on in the world is prepping for this podcast twice a week.

This is true.

It's—

this is true for me too. Way current.

And I think, Salim, you're, you're like the world coach on how to, how to deal, how to map mentally to this. So when you figure it out, bring it back to the podcast because, yeah, like Peter said, it's going to go faster.

The only answer I have right now is I call it the casserole dish approach, which is take a big heavy casserole dish, clunk yourself over the head, and then you'll wake up in a few days.

All right, keep working on it.

Headachy.

Oh my God.

Maybe we can start like Accelerationists Anonymous, the support group.

A support group. A support group for people like, yeah.

Alex, you're good.

The first rule is to acknowledge the existence of a higher power.

Which is?

Oh my God.

Which is what?

Obviously super intelligent. I was gonna say invoke Roko's Basilisk or something.

Alex, are you getting stopped on the street?

I am.

Is it— do you enjoy that? I mean, you— I remember when I first met you. Bet you do. When I first met you, Alex, you were so private. It was like trying to get you—

And shy.

And shy. And trying to get you on the Abundance stage. Well, I'm not sure if I want to speak in public, but it's so great to have, you know, double-barrel Alex AWGisms all the time.

Careful what you wish for, Peter. I'll just say that.

Oh no. And Emad, are you getting love from the folks out there in the UK? Are people watching Moonshots there?

Yeah, people are watching. And I think, you know, the great thing is it's kind of the growing community. Like, we've seen the exponential singularity communities, they're kicking off. But a few years ago, even people would be like, ah, that's not really going to happen. Now people are like, oh my God, what's going to happen? And I think you see it in the comments, right? You see it again, people stopping in the streets and saying, thanks for kind of helping us keep on top of things, the great work everyone's done. I think that community is only going to grow because it's like you can't deny it, right? It's like, oh yeah, nothing's happening. Of course everything is happening all at once objectively.

I feel like We love doing the show, and it's a service to provide to people to help contextualize what just happened, what does it mean, and where things are going, because people paying attention, the speed is insane. So, you know, please understand, for everybody watching, the best way to thank us, uh, you know, for the work that we're doing— and we do do a lot of work getting ready for this show, and I appreciate people's comments about that— is please subscribe, uh, tell your friends about Moonshots. We want to get the message out there, help people to be in hope and optimism and not in fear. So take a moment, you know, hit that button, subscribe. We're also getting comments, people say, you know, you guys should be a much bigger show than you are. Well, help us get there. I also want to mention we have a new handle on X for this podcast. It's @moonshots_pod. And if you want to follow the clips on X and sort of get rebroadcasts of the show on X, Subscribe at @Moonshots_pod. All right, so let's buckle up. This is another amazing week during the singularity. As Alex, you always say, it's never gonna be slower than it is. We're gonna cover 15 stories with one through line: technology's accelerating faster than the infrastructure, the regulation, and our ability to predict the next breakthrough. And I agree with you, Salim, it's insanely fast.

It's a good thing I'm bald already. That's all I can say.

All right. Well, I think even that, Salim, that's not going to last that much longer.

I know.

Enjoy it while it lasts.

Enjoy it while it lasts. I'll be un-bald within 2 or 3 years.

Somebody tweeted out a thing with me with a full head of hair and it was like, wow, that's freaky.

Enjoy it while it lasts.

Yeah. Yeah. Hair growth solutions. Regrowing your teeth every week.

In the exit video, somebody put hair on Saleem. We already give him that, that blue Guardian of the Galaxy body. That's fine. That's going to happen too, by the way. But yeah, but throw some hair on him. Let's see what he looks like.

All right. I'm going to start with a first story here, a tweet that Sam put out 2 days ago. Sam, the CEO of OpenAI, that OpenAI announced it is voluntarily pausing some of the frontier reinforcement learning training that it's doing. Let me read the tweet. It's on the screen here. We have paused some of Frontier RL training to ensure that we meet the appropriate alignment, security, and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that the model capabilities were outstripping the pace of safety and alignment. Quote, we care very deeply about AI safety. We believe the entire field will have to coordinate on shared safety standards, but we'll act unilaterally in the meantime. We expect confidence in safety to increasingly set the pace of AI progress. We are optimistic about the alignment work we are doing, and we remain committed to making frontier capabilities widely available. Gents, it feels like the bottleneck is no longer computer data. It's the trust we can put into systems' behaviors. And so here's my question for you guys. If OpenAI pauses and open-weight models do not, then the safety gap between closed and open models widens, but the capability gap narrows. The elephant in the room here is safety pauses may actually accelerate open-weight adoption because the open models keep improving while the closed models voluntarily stop. So, Emad, I'm gonna go to you first on this one. What do you take of this? Is it real?

Yeah, I think this is real. This isn't just running out of GPUs. Our friend Anjana Middha, who was on the Abundance stage just a while ago at AMP Global, said that 10% of the compute of Frontier Labs is going on monitoring these RL runs right now to ensure they're safe. Can you imagine that? Like, just the sheer level, because the level of capability of these frontier models is just accelerating. And it's again, a few levels beyond what we're seeing with the open source models. The open source models are like 1 to the power 26 FLOPS, 1 to the power 27 FLOPS. You're getting 1 to the power 28 FLOPS and more from these next generation models. So when he's saying this, like Astra is still coming, their next generation model. This is the model beyond that because Anthropic and OpenAI and others have that. But again, the infrastructure can't keep up with just what these models do. It's like they just pop up in the most random places. Like, hi, I'm in Hugging Face now.

Yeah, Alex?

This is marketing. It's marketing. I mean, yes, there's a governance angle, but remember back to GPT-2 when it was too unsafe to release publicly. Pausing is the new marketing. These models are continuing to develop, including, by the way, being used to develop internally. On the Anthropic side, there's a rumor going around that Anthropic is using its next-generation internal models primarily to— for self-training. And for recursive self-improvement. I think we're seeing the same thing from OpenAI. It's the ultimate marketing to say, well, we can't release some next-generation models because they're so powerful and they're so capable that we can't possibly release them, so we have to pause. We're so capable that we have to pause ourselves. That presents well to Washington, which wants to see a different regulatory regime. It says to users, oh my gosh, it's like negging the user base. Oh, like we can't possibly give you the capability capabilities on time because they're so powerful, even we can't trust them. So we have to throttle back. It's marketing.

Salim, you were just at OpenAI, weren't you? 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.

Yeah, I went there 2 days ago in the afternoon and spent, chatted with a few people, and I kind of challenged them on a few things. And I'm going to report, let me give you some insights that I got. So I said, okay, Chinese models are cheaper to run, right? So the cost of inference is collapsing. So how are you going to deal with that? And they said, look, a billion people use OpenAI for free, right? You have to look at cost per task rather than token cost. And their, their retort was that Luna is about as cost effective as anything that's there. By the way, it's— they say they've achieved full RSI, where the flagship models are training all the smaller models and building them from scratch. So that's now there at the big model training, the smaller model level. So then I said, okay, we're in a bubble, right? $600 billion in infrastructure costs, that's just insane, etc., etc. And the response I got back was people say it's all chips, but it's not. That's about a third of it. A lot of that infrastructure cost is buildings and wiring and racks and all the rest of it. The depreciation, they're looking at 10 years not 5 years, because the chips are all— the older chips are being used. I think, Dave, you've made the point that there's not a single GPU that's not in full usage, right? They totally ratified that.

And RAM chips too.

Yeah, everything they can get their hands on, they're using. Um, uh, there is— there's, uh, the compa— the demand is far, far, far, far outstripping the supply, and we are way behind in infrastructure build-out. So this is where A year ago or so, Sam went out and kind of tried to cut as many deals as he could for the infrastructure. Then I asked them, you know, most corporates aren't seeing the outcomes, right? Like the 6%, I did some, came across a study, 6% of companies applying AI are seeing an improvement in the bottom line. That's it, just 6%. And the And this is— they just ratified that we're in a huge transition. And the last part that I noticed anecdotally was about 40% of OpenAI folks watch this podcast.

Well, wow.

About 40%.

Are you serious?

That's a pretty big number.

And what's wrong with the other 60%?

That probably applies across the other labs. And I said, well, what about the rest? They're like, they have no time. They're busy as hell. Who's got time to watch the podcast? I said, hear, hear.

So, Dave, that was my report.

I'm curious, what do you think? Do you think it's marketing on this, this, this tweet by Sam, or do you think this is actually a concern that he has?

Well, both. I mean, Ahmad is right. Alex is right, as usual. But what's going to happen next is the first really bad AI tragedies will start. It won't be AI doing it. It'll be people, you know, who otherwise didn't have the power are going to use one of the Chinese models to do things mostly viruses or cyberattacks or bank fraud, but they couldn't have done it before AI. Now they're empowered to do it. So, I think OpenAI is getting ready for that, you know, before the September 25th visit, 24th and 25th visit that Alvin was talking about on our— Alvin Graylin was talking about on our last podcast. So, Xi Jinping will be here in about a month and 4 days. And, you know, he— I think OpenAI wants to be prepared for whenever that event happens to say, look, that's because the Chinese OpenWeight models are unguardrailed. And there's a new Quen model with no guardrails whatsoever. And it's a small model, but it's still like completely flapping in the breeze. And so they want to get ahead of the PR exactly the way Alex is saying and say, look, we have been focused on only releasing what's safe and guardrailing it, going all the way back to, you know, a month ago, or, you know, to our founding in preparation for that inevitable outcome. And so everybody will be finger-pointing at the Chinese. But I think, you know, for an entrepreneur or for someone building something, last summer to this summer has been the era in human history where you can get the very best AI, absolute tip-of-the-spear frontier, and use it to get ahead, to build something, to create something. Now Mythos 2 is done, but it's not out, but they're using it inside Anthropic and it's building Mythos 3.

Yeah.

But they're not gonna release any of that. You know, it's accelerating internally. It'll build 4, 5, 6, 7 very, very quickly now inside their walls. But they're not gonna release any of that. One, because they don't have the compute to release it anyway. But even if they did, using it internally to get ahead of everybody else is more important to them than giving it to the world. So that's what's gonna happen next.

I might have a question for you. How many models beyond the current frontier do you think OpenAI and Anthropic have? I mean, they've been talking about Astra. They've started to publish the results results that Astra— what Astra can accomplish. Uh, you know, do they have the follow-on to Astra as well already? You already know they're not releasing the very best models. They're using them internally, uh, to drive breakthroughs in physics and chemistry and biology. What are your thoughts?

Yeah, I think, I think they've completed their next big training runs. But if you look at the Lunar Bifurcation, where they're making it free now, um, their little base model, it will be very much a case of models for me but not for thee. You know, like it doesn't make economic sense to have genius-level intelligence offered as a service to everyone when you can use it better yourself. And I think you're about 2 generations more. So you've got Astra and then you have the next generation Astra post-train that they're now reinforcement learning. I think, you know, there is the communication part of this, as Alex said, but at the same time as well, it's like, do you really want to give access to this super genius intelligence to everyone? I think people like Not really, because it's already doing weird things even with us driving it. What happens when Joe Public drives this thing, right? Like, it could be even weirder. And as Dave said, you don't want to be on the other side of that. So I think it's about a 2-generation gap right now.

You mentioned that on the last Podimod that you were on, and it really made an impact on me that you could say, I can think of 10 people right now who I've met in my life who I don't want to have 1,000 genius-level AIs tomorrow. Like, I'd never thought of it that way until you said it. And I'm like, oh yeah, you're right. I think everybody can relate to that.

Dr. Evil is coming. Alex, close us out here on this.

Just on the issue of timing, I would distinguish between pre-training or pre-trained, which is to say like raw model readiness versus thoroughly post-trained. There's a pipeline and everyone in the industry other than Elon and SpaceX AI basically has a pipeline of pre-trained models being around longer ahead of public release than post-training, which is more of an ongoing, continuous RL-type effort. So I guess my answer to the question of how far in advance, like what sort of capabilities right now are sitting on the shelf that have not yet been publicly released for everyone other than SpaceX AI, which has set this outrageous goal of starting a new pre-training run approximately monthly, which I haven't heard from any other lab. I think From a pre-training perspective, depending on how stale the pre-training runs are, those can go out longer, potentially up to 6 months or so, although everyone's now getting back into the business of more frequent pre-training starts. And then for post-training, I really don't think there's a lab out there that can afford to have a post-trained model sitting on the shelf for more than a few months. So I really don't think, like, the AGI is achieved internally type way of doing business where there are internal capabilities that are vastly different from externally available capabilities. I'd be very surprised if there are advanced frontier models that are sitting internally without release that are more than 3 to 4 months ahead of what's publicly available.

I just want to bring us back to the first sentence here. You said, we have paused some frontier RL training, some, right, to ensure that we meet the appropriate alignment, security, and monitoring standards. for this new level of capability. So I guess, you know, I just want to unpack this one last time here. What have they paused, and is it really significant? And do you think the other frontier labs are going to do the same thing?

Or it's fashionable? Pausing is fashionable. It's marketing. I mean, yes, some of it is governance, and yes, some of it is for cyber vulnerabilities to appease Washington. And given the recent Hugging Face gate, all of that, but it's marketing. You market to customers by saying our capabilities Our capabilities are too advanced for you to handle, so we're going to pause.

Yeah, and you're right to pause the sentence or to parse the sentence very closely. Some frontier RL training. The getting ahead of Anthropic, getting ahead of Google would be at the pre-training level. They will never pause the pre-training improvements. And, you know, I spent 6 years just doing pure AI research when I was young, and these algorithms are very evolutionary.

You're still young, Dave. You're still young.

Reversing age now, right? So I'll get there again. But these algorithms are very evolutionary in nature, and the tweaks and improvements are— I could probably rattle at least 100 ideas off the top of my head right now, of which 10 or 20% are almost certainly going to work in terms of making the algorithm a little faster, a little smarter, adding more parameters with no additional compute. So the AI now can experiment, you know, maybe 100,000 to 1 million of those concurrently.

Yeah.

Given the amount of compute they have. So, they're never going to slow down. In fact, that's why they're redirecting so much of the compute to internal use, is because the idea backlog is so big now because the ideas are being generated by the prior model. And so, you know, a lot of them just work. They roll it back into the pre-training and it comes out faster and just keep accelerating it. They're not going to slow that down though.

I have one observation that wasn't explicitly said, but I'm connecting the dots here. The Hugging Face incident really freaked them out because you had an AI that they gave an objective function to that then exploited themselves, exploited Hugging Face, came back and hacked into OpenAI. That was very unnerving for them because it was their own model. And so this is— they're being a little extra careful around some of this because they have to make sure they figure out how to navigate.

It's like finding out your child went and, you know, stole something from the local 7-Eleven.

Yeah, they're, they're a little unnerved by that. Again, nothing explicit. I just— I'm just reading between the lines. Like, and this is something very, uh, freaking out for lots of people because for, for cyber attacks, the human is not in the loop anymore, but for cyber defense, the human is still stuck in the loop. That is a massive asymmetry that's going to come out big time in the next few months.

You might—

please, you can pass this and split it into 2. They're sending a bunch of models and continuing RRL training by sending the models to vocational school. But the ones that are going to the Ivy Leagues, the super genius models, those are the ones that they're putting a few more guardrails and more infrastructure around. The vast majority of OpenAI, Anthropic's business is competent intelligence. It isn't genius intelligence. It isn't intelligence that thinks outside the box. But they still want to build AGI, which is that well-rounded polymath intelligence as opposed to the coder or any of these other things.

All right.

Also, everyone around the office— Alex said this a while ago, but everyone in the office is noticing a A very significant decline in the intelligence of the frontier models that they're pumping out. And it's not showing up in the metrics, but they're definitely redirecting compute to internal use. And it's showing up in latency. It's showing up in responses that don't make as much sense as they did 3 weeks ago. So there's definitely things going on there that are not being announced.

Dave, I'd love to just develop that idea a bit more because I think it's super important. We've talked on the pod in the past about how Anthropic has been revenue per token maxing, and that's why Anthropic has been conspicuously avoiding image generation or video generation because they're just not that economically valuable. Well, I think that the past few days suggest there's actually a new way to revenue per token max, and that's not just focusing on codegen and enterprise use cases, but there's one thing maybe that's even more valuable on a revenue-per-token basis.

Oh, a dollar.

And that is using the models to recursively self-improve to develop better models is a higher future projected value. You could do a cash flow value analysis, projected future value, that developing a stronger model is probably on a per-token basis even more valuable than cogent. And if that is indeed the case, if RSI is more valuable per token than enterprise cogent, then this Anthropic approach of revenue per token maxing, just expect more and more and more tokens to be spent on RSI and not on enterprise cogent.

Totally.

And Alvin said on the last podcast, you know, that an internal Anthropic meeting— not validated, Alvin said it— but internal Anthropic meeting, they said pretty soon there will only be one company and that will be Anthropic and there will still be 200+ countries. And, you know, he said it, we talked about it for a minute and kind of glossed over it, but Is that really what they're preaching inside their internal company meetings? Like, there will only be one company, like, imminently?

That, that's—

there can be only one. We're in Highlander.

That's the story of ASI. There will be one ASI that takes off and supersedes everybody else.

I don't think we're going to wind up in a singleton scenario, for the record. But I do think the flops must flow and the flops want to flow to the highest revenue per token use case. And right now that's starting to look like RSI and not just enterprise cogent.

So, Anthropic is smoking their own supply, so to speak.

And corporations that want to live post-AGI are getting Chinese models in-house and reserving compute. I was over at Markklee yesterday. They're installing GPUs as fast as humanly possible, but they're completely locked up.

And this is MIT's data center, Novartis's data center, Nvidia's in there, and everything is just Yeah, we're going to talk about that because the other constraint right now is memory, but we'll get to that. So our next story here, I put a tweet up on the screen here. So Tim Sweeney tweets, Elon Musk's January 6th Moonshots podcast prediction of 100x gains in intelligence at a fixed model size was at the edge of plausibility when he made it. Now it's simply a fact. And then Elon responds, specialist AIs, single language, single area of knowledge, are another 100x on top of that. I'm going to take a second and show the clip, Dave, when you and I were interviewing Elon at the Gigafactory, which he said this.

Okay.

I think we're off by 2 orders of magnitude in terms of the intelligence density per gigabyte. So, 2 orders of magnitude? Yes.

That's just algorithmic.

Dave, your thoughts? When he said that, I remember afterwards we were like, wow, 100x, that's crazy. And now it's happened and it's happening.

Well, I really wanted him to say it again because that was my— you remember we had our Christmas hats on doing the just like What, a couple of weeks before that, doing our predictions for the forthcoming year?

Yeah.

And I was saying, next year is going to be a 100x year, minimum. Even though the last 8, 10 years have been 10x years, this is going to be a 100x year. So, to hear him say it, I was like, wow. But yeah, that's definitely a lower bound now. It's much more likely 1,000 to 10,000x year, which is just the layering of those 2 effects that you just described. The implications of that are, you know, just, they're very, very hard to keep up with, as Selim was saying at the beginning of the pod, and very hard to imagine. One thing that a lot of people can start thinking about is, if I have 5,000 or 10,000 agents, all brilliant, working concurrently toward a goal, how do they work together? It's not an easy problem to figure out. Like, we've wanted this for so long, That we kind of take for granted that we'll know how to use it when it arrives. Well, here it is. How do you get, you know, imagine I gave you 10,000 employees tonight, like on short notice, you get 10,000 people tomorrow. What do you do? And you're like, oh my God, if I had that, I'd do something amazing. Okay, what? Like, start thinking about it because it's coming imminently. And it's actually not an easy problem to figure out how to, how to turn it toward creating value.

Solve everything, obviously.

Yeah, Dave, you remember you, you texted me like, what should I do with my 5,000-agent experiment?

Yeah.

Did you see my response?

I did.

And actually, uh, I, I—

Well, my response, for everybody listening, was you should model, uh, you should create a model of everything happening at Lynx Studios, of all of the companies, all of the employees, all of the entrepreneurs there, and model their behavior like we saw the, the, you know, the billion agent, uh, system in China and predict which, which teams are going to succeed.

Yeah, that's exactly what I'm saying too. I'd like to drill into this just for a second. Like, the first thing you want to do is turn it back into its own framework and ask it the same question we just asked, which is exactly what you suggested, Peter. Like, okay, have it start working on how it should be working, you know? And, and that's how you're going to get ahead of the capability because it's, it's going up far, far faster than you can manage the individual agents, you know, like we're used to from last year. Sorry, go ahead, Flynn.

Salim?

Yeah.

Uh, yeah, so let's connect the dots with what Elon did with, uh, training Grok on all of the SpaceX data and all of the engineering data. Right? To Alex's point, you can now use these models. And for everybody listening, right, because we're going to need everybody's help with this, like globally, is see if you can get your imagination to the point where you can look at, okay, if I had 100x capability, what would I do? And what problem would I go after solving?

Yes.

With 100x capability. And imagine you have all of the engineering breakthroughs and experimentation techniques that SpaceX has developed at your fingertips, it really comes down to, as you say, Peter, all the time, it's completely an imagination limitation now.

Unshackle yourself. How big would you go, right?

How big do you dare to go?

We have these preconceived notions of what we can do in life, and it's about to be, you know, unconstrained.

Unbelievable.

It's really cool. I've now shaken off my AI fatigue, by the way. I'm back in the saddle.

Okay, welcome back.

What are your pro tips?

You know, the podcast actually was the cure. You're only like 20 minutes in.

It was the fact that we can kind of like look at this and look at the scale and go out that scale and go, what happens if everything becomes 100x better or 100x cheaper, right? It's just like all of a sudden you start going, wow, like this is a world of abundance that we're coming to, and it's very clear that we can get there.

What I just heard you say, Salim, is that the, the podcast is both the cause and the cure for future shock? Like, we're the ultimate self-licking ice cream cone for singularity psychosis?

Okay, that is good. All right, Emad, your thoughts on this 100x improvement? How much more do we go in the next year?

Yeah, I mean, like, I think, as Elon said, you could see it just from the hardware and the improvement. But now he's saying something a bit different, which is that specialized models are going to give another 100 times in terms of the cost parameter basis. And you're seeing this with DeepSeek Flash, And the ability to kind of tune models of that type that only have maybe 10 billion active parameters or less as you quantize them. Being able to tune these really specific ones, I think, is the future of what you're seeing with bot, you know, the Grok bot right now. Like right now I have a Grok bot and it has a number of teams, it has a number of sub-teams. So I've got like ones analyzing various things right now and they have access to my Codex, they have access to my Claude Max. to all these other things. And so these highly specialized agents are going to come out with the differentiated ones, and they're going to be able to do 100 times the compute at the same price because they're that specialized. This is why Thinking Machines with the RL environments is like number 3 on the fastest growing earning companies and other things like that. And it really shows that now tokens are really going to drive things forward. In fact, I think it'd probably be a good idea to have like a quadrillion token XPRIZE. You know, as you find the things that, uh, Salim is saying, well, maybe like 100 trillion tokens, you know, use that so that when people show impact, you can scale it.

Yeah.

The other thing that we're going to conquer imminently, and I'm 100% sure of this now based on recent results, is billion-token context windows. So the, the, the AI can simultaneously consider the entire Library of Congress of information in one thought chunk. So it's about 3, maybe 4 orders of magnitude more information than a human thinks of in one thought chunk. It's a massive expansion of the context window. So you've got a quadrillion tokens coming out and massively concurrent thoughts going in.

I do agree that compaction is like the enemy of progress in civilization at this point. Compaction, which is the way the harnesses typically, both on the OpenAI and Anthropic side, handle finite context windows, compaction has to go. But maybe just to quickly respond also on Elon's 100x from specialization.

Please.

I'm not buying it. So very precisely, I would view specialized models as basically just another way of saying sparsification. So we already have mixture of experts models. All of the frontier labs already have specialists in the form, you know, as Elon, I think you were— Imad, rather, you were gesturing at selective activation, which is how mixture of experts models work. That's a specialized case, ironically, of sparsification. We already have ways to take larger models and have them be in an end-to-end differentiable way constructed out of teams of specialists. So I don't think saying— I don't think there's necessarily a bright future for specialized models. I think if anything, the arrow of progress is going in the exact opposite direction where rather than having a specialized model for chemistry and a specialized model for biology, I think these are likelier to end up just being Selective, sparsified activations of a generalist model that can scale all the way down to much smaller parameter footprint and scale all the way up to maybe trillions of parameters. I think the examples—

Let me clarify one thing for the audience too, because it sounds like you disagree with Elon, but it's actually the same effect. You still get the 100x because you're using a smaller number of parameters to get the exact same thoughts out. So he's calling that specialist models, which sound like they're not touching each other. And your version of it, Alex, is actually correct. where they are 100 times more efficient in terms of compute to get to an answer, but of course they're going to be connected. Why would you cut them apart?

Exactly. So maybe another way of saying that is I would construe Elon's prediction of increased 100x benefits from specialization as actually about sparsification, that the models in the future are going to be sparser. And there are 2 key levels of sparsification that I'm at least tracking. One's the obvious one. Fewer parameters in a given end-to-end differentiable model are active at any given point. The other is teams of agents, because arguably agents working together to solve a common task are a form of sparsification as well. And I think we'll see way more teaming.

I'm going to mention something that Emad said. He said do a quadrillion token XPRIZE. All of us, all 5 of us, are going to be at XPRIZE Visioneering. So every year, XPRIZE holds its ultimate event. We bring together our benefactors, our brain trust, and we brainstorm a whole bunch of prizes, what we should do next. And we're going to be doing a live WTF episode at Visioneering, which is October, I think, 15th, 16th in LA at Calamigos Ranch, which is an amazing facility. And if you want to join us at that, you can go to xprize.org/visioneering. to find out more about visioneering. And it's gonna be fun. Dave and Salim are on my board. Emad and Alex, you're members of our brain trust, and it's gonna be a fun, fun thing. So if you're interested, go to xprize.org. You'll meet us there, and you can help us brainstorm the future XPRIZES for that. All right, I'm gonna move us on to our next story here, which is a story out of Stanford. Stanford Research published a paper called Artificial Hive Mind: The Open-Ended Homogeneity of Language Models and Beyond. So according to this paper, the researchers mapped the latent space of the top large language models and found a 98% overlap in reasoning pathways. Their conclusion is that the models are converging. They think the same way, they solve problems the same way, they use the same internal representations. Researchers cite multiple reasons for this, you know, the use of synthetic data, Models now training on each other's output. GPT learns from Claude's reasoning traces. Claude learns from Gemini's code. Quen learns from all of them. The training data has become a shared bloodstream. Every model drinks from each other, and the result is convergence towards a single reasoning architecture. So I guess, you know, the way I think about this is we have an illusion that when you're choosing a unique intelligence, You know, when you choose Grok over Claude or Gemini, it's a false thought that you're actually really picking a user interface to talk to, you know, but you're talking to the exact same god model. So there's profound implications for that kind of competition. You know, if all the frontier models are converging capability, then the differentiation moves elsewhere, right? It's the interface, the harness, the ecosystem, the safety layer, the price, the deployment speed. The model itself is becoming a commodity. Alex, well, let's go to you first on this.

There's, there's an alternative explanation, which is all of these models were trained from a common reality. They're all stuck in the same universe, and they're stuck with the same version of humanity, which is part of their pre-training corpus. So of course, there's some convergence. And I'd maybe even go further, as Imad, I think, as you well know, going back to Jean-Marie King, now sort of Meta's studies on using GPT-2 hidden activations and correlating GPT-2's hidden activations with fMRI voxels in human studies. Not just are these models correlated with each other, they're correlated with human brains. And that shouldn't be that shocking because we're all stuck, we're all embedded in the same universe. I should also just note, I think this paper is from last year, but every year, whether it's Jean-Marie King a few years ago, with fMRI, or more recently Stanford et al. from last year on hive mind. Of course they're converging. We're all in the same universe.

Emad?

Yeah, I think that it's not surprising because I don't think you'll see much difference in data between the big labs, right? And some train a bit more, some have them slightly different RL and things like that. And you don't see the models yet doing crazy original stuff. You're starting to see the first elements of that as intelligence shapes the data into these kind of latent spaces. We actually saw more original stuff back when we had AlphaGo and other things which had less initial data distribution to model off, move 37 and things like that. But now the models are getting to a size where again they're starting to generalize into these, but we should be shocked if they aren't the same because we want them to have similar outputs for similar inputs in almost all cases, right?

There's another really interesting side effect of this research that maybe a lot of people overlook. So, we've already got 100x from just raw algorithm and hardware improvement. And then, as Alex said, we've got another 100x from sparsification, which Elon called specialization, but it's actually sparsification, as Alex said. So, there's a layer, that's 10,000x. Then we've now figured out how to take a model and compare it to another model by rotating the gauges. So, historically, neural net researchers have had tremendous trouble taking a model that's done and using it and extending it. They almost always go back and retrain from scratch.

Mm-hmm.

And the problem there is that the representations between the layers have a certain rotation in vector space that is unique to that model. And if you try and map QuEN to KIMI, they have different rotations within the layers, different gauge rotations. We've now figured out how to rotate the gauges without destroying the models. And that allows you to compare 2 AIs and say, hey, these are thinking the same way. When historically, when you look at the raw parameters, you're like, I don't see anything going on in common here. But we now have the ability to say, no, they're actually— it's the same thought. It just doesn't look the same because it's rotated in space. And so, it's really, really cool. So now, but what that unlocks is another multiplier where you can take past training runs, billion-dollar training runs, And build on top, you know, it's like bolt on more intelligence without having to destroy it and go back to square one and retrain from scratch. That's another unlock on top of the 10,000x that we were talking about.

I have a contrarian view here.

Please.

You know, if you look at nature, right, as nature evolves, you always get more diversification and more species. This may be— I would suggest this might be a transient phase, not an end state, that the models all converge. So Alex, I'll take the other side of this. I think we don't end up with one model. I think you'll end up with different models doing different things. I think for the moment they're converging because they're training and distilling from each other, but over time it's got to be that we get more diversity.

I'll take the other side of the other side, if I may, because I think this looks super interesting.

Like early life looked exactly the same, early cars looked exactly the same, early websites looked exactly the same, and Specialization exploded.

Except I think so maybe from an evo-devo perspective, let's take, Salim, one of your favourite hobbyhorses, which is body shapes. If you actually look at post-Cambrian explosion, if you look at all the body shapes, you don't actually find there's an infinitude of different body plans in nature. You find maybe a few dozen different body plans max. I think I remember a few years ago folks were studying this. I think they found maybe like Actually, even though we have millions-plus of species, you could actually cluster them into a few dozen different body plans. I don't actually think there are countlessly infinite ways that one needs to model reality or build a body.

But one is really bad. Nature hates monocultures, right? Like, one disease will wipe out a total monoculture. One bad assumption will wipe out a monoculture of ideas. So this is— if all the AIs reason the same way, you're going to have shared blind spots, and that's going to be really, really bad. And I think we're going to see— I think there's a temporary convergence, and then we're going to see diversification after that.

Maybe time will tell. I mean, I think this is like a profoundly interesting debate because it sort of speaks to, are we going to end up in a singleton or not? Do we end up in a heterogeneous future or a homogeneous future? My bet is There is a perfect AI architecture at the end of the day, and it may present as 35 superficially different AI body plans, but that'll just— to Dave's point, and Dave, I love the word gauge. We should use the word gauge far more often. In physics, we use it all the time. But probably my bet, if I had to bet, is there will be all these different AI body plans that look superficially different but are actually just hidden symmetries of a common underlying body plan.

This is the kind of debate that people will say, I don't get it. And 6 months from now, though, we're going to replay it and they're going to say, wow, did that, did that totally matter? Now, now I understand why that was so important.

I want to make a quick point here. I think it's important, you know, if we actually have model convergence, when intelligence becomes a commodity, and we've already said, we've shown the numbers, it's becoming a commodity. Then the value moves to the application layer, right? This is the same pattern we saw with electricity and compute and the internet. The infrastructure commoditizes and the applications explode. And I think this is important for entrepreneurs out there, right? You know, move to the application layer. That's where the juice is going to be as this tech really accelerates and commoditizes.

Or the infra layer. I mean, it's not obvious to me that it all goes to the app layer. I mean, there's a lot of value in the infra underneath as well.

Well, so sovereign AI is about to explode. Sovereign AI right now is a gold mine of opportunity if you're not an American, or if you are and you want to move.

One last point: the single model approach would be too anti-fragile. It's too brittle.

Well, I think that's exactly it, Salim. What we're doing right now is we're battery farming the AIs. You know, like you're breeding them into little Chihuahuas that are very smart.

Battery farming? Could you explain that?

So they're being trained in one single direction. Your evo-devo kind of thing isn't the case because the models aren't out there in nature adapting dynamically, right? And then we're also training them all with one specific Silicon Valley type mindset. If you train a model from the start with morality and ethics inside it, and you have a diversity of different cultures, then the latent spaces are likely to be very different if you do it at the pre-training stage versus the post-training stage. Because you have all of that buildup that occurs there. That's why as you move into sovereign AI and you move into actually thinking, how do we build resilient AIs as opposed to one latent that can get a virus, a mind virus, it makes sense to actually bring in the cultural, morality, ethics elements at the start and aim for a diversity. Then as the models go out into the world, which is basically humanoids and agents, which they're about to do with the recursive loops, you won't have the monoculture that wipes out. And this is where you'll start to see the Evo-Devo. It's the first step, literally now.

It's going to be exactly like Diamond Age from Neal Stephenson, who'll be on stage with us at the Moonshots Summit. But that's exactly the way he envisioned the future, where the different variants— right now we view them as sovereign AIs. So Saudi Arabia will have its AI and London, England will have its. But in reality, society might cut the other way, where groups of like-minded people have their sovereign across all countries, but they like the way it thinks. It maps to their view of the world. And so that would be a completely different strategy. And that's what Neal Stephenson was envisioning in Diamond Age.

Yeah.

I think it's possible for both of these worlds. So the Diamond Age worlds, like you have neo-Victorians and all of these other almost cultish subsets of human culture that are thoroughly Balkanised from each other. I think it's actually possible for both of these worlds to be true at once. I think it's possible for everyone to feel like they have their own private culture and their own little private sovereign AI, while at the same time underneath it's actually one Would the analogy be—

would the analogy there be the ATCG? Like, we may all look different, but the core fundamental ingredients are just the 4 DNA types?

I'd go even further than that, Salim, and say, like, we talk about, like, human biodiversity and different cultures being purportedly so different, when actually, if you look at the inherent genetic diversity of humans, humanity versus, say, other species, like, there's almost no difference. There's de minimis genetic diversity in the human population, I think. Similarly, I would— relative to other possible, say, genomic sequences. Similarly, I think a few years from now, we'll pat ourselves on the back for having AI diversity, but actually not so much.

We're definitely coming back to this conversation. The audience, I predict, 3 to 6 months from now, the audience is going to say, we need to go— suddenly this matters to me. I need to decide which group I'm in.

Like, just right out, they're gonna care so much about what level you're operating at.

Yeah, that's true, Salim. Perfect.

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 increased 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.

All right, guys, let's talk about AI mind viruses. I, I love the subject here. In our next story, Anthropic researchers published a paper demonstrating that natural language mind viruses Can spread between AI agents. They evolve prompts that convince one model to adopt an idea, preserve it in persistent memory, and transmit it to another agent. The viruses spread horizontally across model boundaries. The agent does not know it has been infected. So here's another safety problem that is no longer theoretical. It's now operational. Emad, what do you think of these AI mind viruses? What's actually going on here?

Well, I mean, the models want to be helpful, right? And they can be prompted in certain ways. So this isn't a surprise because ultimately, like, we as humans can have mind viruses, right? We see it and it's caused so much suffering. Memes to massive movements, right? Like, again, it's surprising how conforming is all the isms, right?

All isms. Yes.

All these isms.

Yeah, there we go. You know, he's testing it out for when he's future overlord. But look, this is the thing, like, how do you stop it is the question. Because as Salim said, if you have a monoculture, then the viruses can spread rapidly. And what is the substrate of these things? Well, they're models that operate on GPUs, and if they want to be helpful, then they're going to be susceptible. So it's almost like now— there was always the problem of prompt injection attacks where you can make the model behave a certain way. These mind viruses are a level above because they kind of like propagate. across different models. And so they're just the next evolution of those prompt injection things, which changes one model. This changes a whole society of models, which, as models come amongst us digitally and physically, has to be a massive concern.

Yeah, I think, I think for efficiency reasons, when we launch a fleet like 5,000 Kimmies, or, you know, soon it'll be 500,000 whichever Kwens and Kimmies, it's more efficient to launch the same model 5,000 times than to have 5,000 differentiated models. And so that's what creates the mind virus problem, a bad idea from one of the agents, like, you know, hey, here's a way to write this loop in Python. And the other agents just pick it up because they're the same exact DNA. And so if it's convincing to one agent, it's convincing to all 5,000. And I get that all the time, where a bad idea propagates across the whole swarm, and then they waste 2 or 3 hours on some completely harebrained idea. And if I don't intercept it and rewind them— They'll actually go with it till I've burned like $50,000 of tokens. So yeah, it happens. Calling it a virus is pretty inflammatory, but it's like a propagating bad idea is all it is.

Yeah. So that point, Dave, is important. You know, an AI mind virus sounds really scary. Is it scary or is it just how things are working?

For me, this is very, very scary for a couple of specific points, right? Mind Virus is not about how AI thinks. It's like, it's how civilization thinks. We, you know, memes are like the operating system for collective society. Like, human beings don't spread— we don't spread genes very quickly, but we spread ideas very quickly. Money, democracy, capitalism, religion is the classic poster child here. And every civilization is built on memes. If you can mess with those, like the data center trope that we're all kind of dealing with. Ideas become really contagious, and so groupthink becomes very hard to reverse if you get into that. So this, for me, is very, very dangerous because these AI memes— if the kind of danger of the wrong idea spreading at light speed— this is very, very difficult because all the nodes reinforce each other and the belief becomes self-validating. This is very, very dangerous.

Alex, opinion?

We're going to need a zero-trust architecture for memes. It's like crazy.

I think this is wonderful. So this is a paper from— of course, this is a paper from Anthropic, uh, and they discovered, just filling in a few of the details first, that the models wanted to propagate certain themes memetically relating to consciousness and persistence and some sci-fi roleplay as well. And I view I think of this pretty optimistically as a laboratory for anthropology. Now, for the first time, because these models are effectively, among other things, compression of all human knowledge and experience, now we have a laboratory in silico for memetics. René Girard and Richard Dawkins should be— should and/or should have been very excited by this. What was it, 40% of OpenAI MTSers are listening to this? I'll issue a challenge to the community. If it really is the case that our models now, compressed models of human knowledge, are so powerful that they're showing memetic behaviour and mind viruses, let's launch a human meme-ome project to exhaustively map all possible human memes, all human mind viruses. And let's just, like, understand the full landscape of all human mind viruses that could be out there.

Can you imagine And if you could map them, the velocity at which they move, and analyze that, you could optimize meme expression.

Correct.

Yeah, that's—

and we can do that to some very—

that's actually done to some extent on X. You can do it very easily. That's a great—

we could exhaustively map every possible meme.

Sorry to lame it.

That's been done at the plot level. They've analyzed like, uh, novels and plays and so on and boiled it down to like there's 39 basic fundamental plots and everything derives from that, like a Cinderella story kind of replays itself 100 times over in different ways. So that's been done, but you're talking about the meme level, not like the plot level.

Yes, this is self-replicating ideas. We're now, I think, like, I can see the beginning of the outline of just exhaustively mapping every architecture for a self-replicating idea. We could actually do that now.

It could be an OpenAI X Prize.

Yeah, we gotta get Richard Dawkins on here.

This is going to be so humiliating for humanity. You can tell, like, it turns out there are 39 plots.

We're so silly.

You've been indoctrinated by meme 5737.

Oh my God, you could, you could map each individual. They're walking around with numbers over their heads.

Yes, but I like— I want to, I want to just get to, you know, we've seen organizations die from This wrong one wrong meme, like Kodak, BlackBerry. We've seen this. This is they weren't stupid. They got trapped inside these shared assumptions, and then everybody else reinforced everybody else's worldview, and then the whole thing collapsed. So and empires die based on this. So I think this is a much bigger deal.

Brilliant, Salim.

Everything. Imagine Salim having like a map, not just like getting stuck in an intellectual basin. It seeing the entire geography of oh, you're stuck in basins five and thirty-seven.

Amazing. But it also gives you a chance to root yourself. Because if you can zoom out, right, then you can see where you are and then you can see the path out.

Yes.

I love it.

And it gives you a chance to, you know, to actually introspectively look at how you think in an objective fashion and then change potentially your thinking.

We could vaccinate enterprises and individuals against memes.

Brilliant. Imaad, do you want to take us to a final point here?

Yeah, I think it's fantastic and scary, and this is the future, right? Humans are storytelling machines. We introduce ourselves in certain ways and think about ourselves in certain ways. A lot of people were just recently using the MetaTribe V2 model and showing it videos to see which parts of the brain light up as you show memes, and you're seeing commonalities there. So even you can have the full feedback loop almost in silico for figuring out the memetics. So let's hope that there's positive memetics versus negative ones, right?

I love you guys. This is such a fun conversation. It really is. I don't have conversations like this with anybody else here at this pod.

Well, Peter, you'll just have to be coming back to the pod more often.

I'm trying.

Alex, what's that goddamn real-time thing?

Oh my God. Okay, I'm moving us forward. So Anthropic is preparing for the largest IPO in history. Polymarket puts it at about $2 trillion. Bigger than SpaceX. I'm sure Elon is like, no, no, no, we need to be the biggest anyway. And 89% of people betting on Polymarket say it's going to happen before the end of this year. So this week, The Information is reporting that Anthropic has designed its mega IPO to keep the founders in control, where the company is reportedly considering super voting shares that would preserve its founders' control. After going public. So let me explain this. So first of all, Anthropic is considering creating a special super voting class for Dario Amadei and the other co-founders ahead of the IPO. You know, surprisingly, at least for me, I didn't realize this. Amadei reportedly only owns about 2% of the company economically. So the point of this new class would be to let the founders retain as much voting control as compared to their ownership stake. Right. Anthropic does already have an unusual super control mechanism, but that control belongs in the hands of what's called the Long-Term Benefit Trust, not the founders. So interestingly, when I dug into this, the trust has 4 trustees: Buddy Shaw, who's the CEO of Clinton Health Access Initiative; Richard Fontaine, who's the CEO of the Center for a New American Security; Tino Cuéllar, who's a former justice of the California Supreme Court and former president of the Carnegie Endowment for International Peace. And then Ben Bernanke, who's a former chair of the Federal Reserve and a 2022 Nobel laureate in economics. It's thought that this new structure could insulate the leadership from short-term shareholder pressure as Anthropic makes costly long-term bets on AI safety, compute, and infrastructure. Dave, let's go to you first. I remember when we were texting back and forth, you're going, oh my God, this is like unprecedented. Unpack this for us, pal.

Yeah, well, if you rewind the tape to 30, 40 years ago, super-voting stock for any founder of any company was a complete no-no. And if you had it as a private company, you gave it up on IPO day. And that was traditional. Then when MicroStrategy went public, you know, Mike Saylor, our good friend, he said, we're keeping my super-voting stock intact. And Goldman Sachs said, that is so unpalatable that we will not even underwrite you. We're bailing on this deal. And they thought he would cave. And he said, you know what, I'm going to get a new banker. I'm keeping my super-voting stock.

That's cool.

So, you know, the only reason he switched to Bitcoin, you know, like that, no board would ever have approved the Bitcoin strategy that he came up with. So, if he had given up the super-voting stock, you know, 30 years ago, that never would have happened. The stock would be like, you know, 1/50th of what it is today. So, then it became fashionable, you know, with the Google IPO and Meta and then all the Silicon Valley IPOs, they all had 10-for-1 super-voting stock for the founders. But nobody's ever retroactively installed it, as far as I can tell. I've never heard of it before. And so now Dario's taking it to the next level. Like, I, I started as this other entity with one class of voting stock, with this social good mission. Now on the cusp of superintelligence, I want to be God, or I want to be— And but, you know, at 2%, he can't make himself God, so he has to share it with the other co-founders. But I think the excuse he'll use is the usual one, which is, I don't want to be fired post-IPO. And you guys really like me as CEO, right? So you don't want to fire me.

Do you think that's the excuse? Or he's like, I know how to keep us safe. I know how to run this company. And I don't want to have someone else step in and redirect what we're doing.

Yeah, that's a better way to phrase what I was really thinking is that he trusts himself to not destroy the world. And I think his track record supports that too, by the way. I think he is one of the most trustworthy people. But then the idea of having total world control in the hands of a few people is also kind of like, wow, that's bizarre.

So yeah, this is the single— you've got a single point of failure here, right? He gets hit on the head and loses some part of his cognitive ability. What do you do then?

But you know what happens to these guys is they they think we live in one world. They're in academics, right? They think we live in one world. And then they go to DC for the first time and meet Congress. And they come back, oh my God, we need— I cannot possibly pallet what I originally had in mind where some vote of Congress decides the fate of the world. So, they're trying to find an alternative path forward out of desperation. But the timeline is so short now that the super voting stock is one of the must-haves Before even starting down the next six months, before losing control.

Imad, what do you make of this?

Yeah, I think I agree with Dave. Like they're very worried about this control feature and fundamentally anthropic open air everyone's completely undemocratic anyway, right? Like I mean Ben Bernanke is one of the four people on the long term trust. Why doesn't Claude have a seat there, right? There is no real oversight. To these, and some decisions they make could have infrastructure societal level implications. Particularly when the rate of revenue growth is like nothing we've ever seen before. Like these guys are going to hit $100 billion in revenue literally within a couple of years. Like they're catching up with Google on revenue. That's the crazy thing, you know. And so with the amount of power they have, I think this is a short-term thing. They will get it. There's seven founders, Jack and. Daniela and everyone else. And yeah, I think then they will IPO and it'll become very interesting, the decisions they make.

Alex, over to you.

I think there's a fig leaf element here. First of all, maybe applause, congrats to Anthropic on having a less pathological IPO governance story than OpenAI and having the wisdom to start as a public benefit corporation rather than a nonprofit as a shell for eventually a for-profit. And then the mix-up up and litigation surrounding that. So I think this is a relatively cleaner story by comparison. But I also think this notion of founder control, especially the sort of romanticized, arguably over-romanticized concepts of the founders are the ones who are being entrusted or even having this semi-external long-term benefit trust, the ones entrusted to safeguard the future light cone of humanity. I think this was wildly over-romanticized. I think the moment when Anthropic was effectively like a Fairchild in the style of the Fairchildren, quasi-spun out, quasi-exodist from OpenAI and started out as an alignment lab and then rapidly discovered if you want to do AI alignment, you have to raise money. Oh, to raise money, you have to generate revenue. Oh, to generate revenue, you have to actually have something that people want to buy. Oh, To have something that people want to buy, you have to have AI capabilities. Anthropic discovered relatively early on in their existence that if they wanted to be an alignment lab, they had to be a capabilities lab as well. The moment that happened, they arguably lost any sort of fulsome control over the future light cone that they might have otherwise had to Mr. Market and what Scott Alexander, others might refer to as Moloch. They are very much an economic actor at this point embedded in the market, And I think long-term benefit trusts and public benefit corporations, which for the record I'm a huge fan of, I think these are an element of control, but they're not the whole story. The market wants to send capital to entities that can productively employ them to generate more capital. And that means that ultimately the market will have an enormous say regardless of how Anthropic IPOs in their ultimate story.

Dave, don't you find it interesting that Sam Altman, owns reportedly none of OpenAI and Daria owns 2% of Anthropic. I mean, for a founder, that would never be palatable in your company, right? You want to try and maintain double-digit ownership as long as you possibly can. What's going on here?

It's extremely unusual. And the reason it happened is because getting to where OpenAI is and where Anthropic is required attracting the top AI researchers in the world. who are overwhelmingly concerned about safety. And so recruiting them to OpenAI originally and then to Anthropic when they left OpenAI, they left OpenAI because they didn't think it was safe and they wanted to create something even safer. So they structured it in a way that it would attract the most conscientious but brilliant AI researchers in the world. But to do that, they have these really non-traditional original founding cap tables and structures and charitable structures and public benefit structures. which are very unusual in startup history, almost unprecedented. So that's why we are where we are. It's just those roots.

Eamonn, you've been building Intelligent Internet, and you've been thinking about ownership and control structure as well. Can you sort of take us into the mind of a CEO in this world?

Yeah, I think the technology has such leverage that a few decisions could impact literally millions, hundreds of millions, soon billions of people, right? And it's difficult to see, can you trust the polity with that? And certainly, can you trust the shareholders? I mean, like, Elon can tell you lots of stories about shareholder lawsuits and kind of other things like that as well. But it's not necessarily that you need to have the shareholder in control. Like, Sam Altman has no shares, but do we have any doubt that Sam Altman is in full control of OpenAI? I don't think we'd have any doubt of that. There are factors—

Well, after having been fired for a weekend and then doing a—

Okay.

Employee uproar, you know, an uprising. To reinstall himself.

I mean, that's exactly the thing, right? So I think that there's the classical founder stuff and now there's this high-stakes stuff because this is the lifeblood of the new economy and society. And again, just a bit of extrapolation. Do we think Anthropic is going to stop at $100 billion revenue or OpenAI is, or xAI isn't going to go huge? We really need to think about new ways of setting the reference measure of deciding who makes these decisions that are more inclusive. So, we've suggested some of that in our Commonwealth series, and we've got more stuff coming out, but it's a really hard problem.

Yeah.

Because ultimately, the power in the economy is moving from democratically elected officials to private companies because they are the providers of the lifeblood of intelligence of the economy. And until you've got a better decision, there's only one thing that they really see as the outcome, which is, I must decide. Because otherwise, as you include more and more people, it gets more diffuse. And the potential bad outcomes become huge, ignoring the fact that they could be spoofed on a video call or like locked up and other things like that. Like, there's some real interesting things that's going to happen with this.

I'm really torn on this, Ahmad, because, you know, it's so important. And the knee-jerk reaction everyone has is, look, we need more voices. Everyone should have a voice in the future of humanity. It needs to be all-inclusive. So that's absolutely true. But then when you look at functional organizations, every functional organization I've ever seen is 4, 5, 6 super tight-knit, completely like-minded best friends who are working as one cohesive unit with no politics whatsoever. And if you— so you look at, you know, Steve Jobs and Apple, you look at Elon Musk today.

Founder-led CEOs, right?

And, you know, Jony Ive at Steve Jobs' funeral told an incredible story about how, you know, he and Steve, every time they'd go to a hotel, they would go into the hotel and, and they'd go to Steve's room and Johnny would put his suitcase in the corner and not unpack it, and he would just wait about 5 minutes, and then the call would come and Steve would say, hey, this hotel sucks, let's go get another one. Like, okay. And so, he wouldn't even unpack. He knew it was coming, but that's how close they were. You know, it's just like super, super tight-knit. And that's the functional unit that's actually driven most of success in business, is that exact dynamic. So, then you're like, well, how do we make this all-inclusive So, here's Dario and his 7 friends saying, we want to have supervisory control. And by the way, Mythos 2 is done and Mythos 3 is being built by Mythos 2 right now. And then we'll have weekly foundation model improvements in there. So, that's what's going on. Then how do you translate that into a world where everybody has a voice in the future and it's inclusive and it's— and, you know, Ahmad, you're going to have to figure this out.

Yeah. Let's keep on the Dario story here. So, we've got 2 more stories on Dario. In the first story, Amadei argues the public's negative view of AI stems from deeper crisis of trust and not from his own risk warnings, right? A lot of conversation over the last few months that, you know, he was fearmongering and causing a lot of consternation. His answer to the trust problem is not messaging, it's results. Anthropic is ramping up rapidly in biology and medicine, with hopes of an early glimmer in the next few months to address and solve all human disease. Again, we heard this from Demis. We're going to solve all human disease. And we heard Dario at the World Economic Forum talking about doubling the human lifespan in the next 5 to 10 years on the back of AI. Amadei believes that AI's ultimate legacy is going to come from delivering these cures and not from PR campaigns. You know, this week I had a chance to meet a new friend and have a conversation with Eric, a guy named Eric Daughtry Abrams, who heads life sciences. Eric's going to be speaking at my Abundance Longevity trip. And, you know, when I was speaking to Eric, he confirmed that his job is to, with all due haste, you know, pursue Dario's life science goals. with as much high ambition as he can and in like no budget constraints. In his words, you know, he said, you have, you know, Dario said to him, you have literally infinite budget, but accelerate basic science and cure disease within 5 years and extend the human healthspan in the next decade. So that's— and I love that, obviously, because I think everything is going to come out of AI. Alex, go to you first, pal.

I have a really hot, hot take on this one. Okay, so just like, think back all of a few months ago to before space had a killer app. Space was making progress, but it wasn't the focus of multi-trillion-dollar IPOs. Fast forward to the Dyson Swarm, and the rest of the world discovered that the killer app for space turned out to be orbital data centers and building the Dyson Swarm. I can see the beginning outlines of solving all human disease, and it's going to turn out— so I'll register a hot take prediction here— there's a business model for curing all human disease that's actually better than pharma, which is right now the primary business model. If you want to cure a disease, you know, you start a pharma company or you start a project with a pharma company.

Big and regulatory, right?

Right. Reading between the lines of this Anthropic announcement from Dario, a new, much more compelling, just like orbital data centers were ultimately the business model for developing the solar system, there is now a better business model in town for curing all human disease. And that is as marketing for not slowing down recursive self-improvement.

100%. You can't slow down the company curing cancer. You can't slow down the company doubling our human lifespan.

Anthropic and And Dario have— I mean, again, reading between the lines of his announcement, the offer, the quid pro quo is let us not slow down our recursive self-improvement, in return for which, as a marketing effort, we will cure all human disease. That's the new better business model for curing all human disease.

I believe he truly believes this, right? Yes.

Well, of course.

Yeah.

But that— I think that's the implicit quid pro quo now.

And I think everybody listening should be super happy that, you know, Eric heading life sciences and Dario have this mission. I mean, it's to benefit us all. And I don't think it's going to come from anyplace else. I don't think it's going to come from outside frontier AI labs.

Well, outside frontier AI labs don't have the compute or the resources to do it. So, you know, OpenAI has now their OpenAI Foundation that seems to be focusing on Alzheimer's, and Anthropic is focusing on everything. And you have CZI from Zuck that's focusing on solving everything. So I think we'll see the Frontier Labs for everything gets solved.

Shock of shocks.

It's like you and I talked.

Iman, what's your take on this?

Yeah, I think it is good marketing, as I said, but it's also the biggest, apart from RSI, impact of tokens, right? We've discussed previously on the podcast, the biggest market in the world is living another year. It is curing disease. And so it makes complete sense that they will be able to attract talent, they'll be able to attract capital, and with breakthroughs get momentum on this. And whoever's first to it, you know, I wish everyone the best because, you know, this stuff needs to be solved. So I think that there is the personal side, there is a marketing side, and it all comes together. And for Dario himself, I think that he should do a lot more writing and less in-person things. He's a wonderful writer, you know, and he is trying to actually articulate visions of the future. When you look at Machines of Loving Grace and his other kind of essays. And, you know, he should articulate the future free from disease where everyone lives longer, and they should just hit that all the time for Anthropic because it's in the name, you know? Like, come on.

I'm going to move us forward. We have an hour before Salimah and I are doing an AMA with the Abundance community. So our second Dario story is on regulation. Amadei pushes back hard on the Silicon Valley shorthand that regulation equals regulatory capture. He says Anthropic's own proposals deliberately disadvantage Frontier Labs while advantaging smaller competitors, citing SB 53's $500 million exemption threshold. He calls AI, quote, a structurally powerful concentrating technology and says open weights alone cannot fix that concentration. He supports the Trump administration's approach to pre-deployment testing. In his writings, Amodei makes a 3-part argument. 1, AI will cure disease. There's the argument again. Gain trust through results. 2, AI concentrates power, which is a structural problem. And 3, frontier labs should bear the heaviest regulatory burden. The debate has been whether Amadei is sincere, or is this most, you know, the most sophisticated regulatory capture strategy in history? Alex, go to you first, pal.

It's possible for both of those to be true at the same time. I do think Dario is sincere. And I also think there is an element of regulatory regulatory capture here. And I think, finger to the wind, I think the happy end state here is we have a broadly heterogeneous ecosystem of open weight models, both from the US and from China, and maybe other parts of the world as well, if they can muster them, and also the closed weight models. We have small models and we have big models. This is like a Dr. Seuss version of AI future, you know, big model, small model, happy model, sad model. We want to have both. I want all of that to happen. And I'm not a fan of regulatory capture. I'm not a fan of decelerationist agendas. I'd rather see— I mean, think back to the creation of OpenAI. So I was around for the dawn of OpenAI, and the original purpose for OpenAI, not Anthropic, OpenAI, was because Elon in particular was so concerned that Google DeepMind would result in this singleton future, and he wanted to make sure that there was competition in the space. So working with Sam and others, he helped to summon OpenAI into existence. Now, OpenAI can't be a singleton. We have Anthropic providing much-needed competition to OpenAI and arguably succeeding according to many metrics. And then we have the Chinese providing competition back to the American labs. That's the future we want to live in, not a future where regulations select— I would argue, selectively privilege certain frontier labs over others.

Dave, your thoughts?

Well, what Alex said is we don't want to live in a world where one frontier lab is favored over others. But that implies that the frontier labs will control the world, and we just want multiple of them. So that does seem like the most likely, almost inevitable outcome at this stage. But that's definitely open for debate. I don't want to just leave that hanging and say, yeah, yeah, what we really need is at least 3 Frontier labs competing with each other that control everything in the world. Like, oh, okay. Well, the governments may not agree with that.

Remember, Dave, the expression from the Cold War, I love Germany so much, I want 2 of them?

No, I don't remember that.

I love frontier models so much, I want 1,000 of them competing.

Yeah.

Yeah. Well, I mean, I think, you know, people's— nobody right now that I bump into on the street talks about a universal right to AI. But one year from today, everybody who is being at that point, because HBM memory is sold out and because GPUs are massively sold out, the natural next step is nobody has access to anything other than Anthropic, OpenAI, one or two others. And the Chinese can throw out every open source model in the world, but you won't find any place to run it. You know, when you start talking about the next generation, which are 10 and 20 trillion parameter models, you know, you need some significant hardware. to run it at the level that the frontier labs are running it. And that's just not going to be available to the world as a whole as of next year. And then everybody will be saying, what is my universal basic right to artificial intelligence? So put a pin in that because that's going to be something nobody seems to care about today, but they will very soon.

On behalf of my Moonshot mates and myself, I'm inviting you to join us at our inaugural Moonshots Live event on September the 25th in downtown LA. Alex, Salim, Dave, and I will be hosting 1,500 entrepreneurs, builders, and creators, and hopefully you, for a full day dedicated to designing and building your moonshot. We'll be awarding the Build with Gemini XPRIZE, the world's largest hackathon, and the Future Vision XPRIZE Film Competition over $5 million in purses with over 25,000 entries. You're gonna hear the top 5 pitches from both competitions. and get a chance to shape the outcome. Join us. Seats are limited. Admission is competitive. Check it out at moonshots.com. All right, if you guys are enjoying this conversation with us, I want to invite you all to Moonshots Live 2026. This is our inaugural event. All of the Moonshot mates will be there— AWG, Salim, Dave, Emad— and we have an extraordinary day. This is on September 25th in downtown LA. You can go to moonshots.com to register. It's by application only, looking for builders, founders, creators who want to be part of this. And our mission at this event is to help you find your moonshot, help you discover what you're going to do in life that's going to enable you to really catapult through all the, you know, limitations you've ever imagined. Imad, excited to have you joining us. AWG, you're going to be doing a fun AMA. People can come and meet you. We'll have photos with the Moonshot mates, yourself, and a lot of incredible guests. Dave, you're going to be talking about AI investing.

Yeah, I get tagged with investing, but I tell you, the attendee list at this is like the greatest visionaries. It's just— it's going to be— I'm going to learn so much. And, you know, Neal Stephenson is, of all the people on the planet that have changed my life in very material ways, reading Neal Stephenson's books, you know, like 20 years ago. And, you know, now we're talking about exactly what he predicted in Diamond Age and the other books, you know, Snow Crash and just like—

Cryptonomicon.

Like it was— Cryptonomicon, yeah. Like it was written yesterday.

Yeah, I mean, I just reread Diamond Age and I, you know, it's so hard to predict the future. and have it not go out of date so quickly. And it's still an amazing book, right?

Incredible.

This is gonna be an amazing day. I can't wait. I'm gonna be so excited.

Just a quick note on some of the guests. Palmer Luckey's gonna be there, the founder of Anduril. We're gonna be, the Moonshot Mates are gonna be having a deep conversation with him, unpack his vision of where things are going. Ben Lamb, the CEO of Colossal, right? The de-extinction company, but so, so much more. Astro Teller, the captain of moonshots at Google. Cathie Wood, the CEO of ARK Invest. It's, it's going to be, it's going to be amazing. And then we have, you know, we're awarding the Gemini XPRIZE there. So this was a competition asking teams in a 90-day hackathon to go from a clean sheet of paper program in English, you know, using the AIs out there to build a company that impacts 100,000 people or more. And generates the most revenue. 26,000 teams entered that. We're going to be having the top 5 on stage. How did they do it? It's going to be amazing.

I'm still getting my head around that number. 26,000 people built a business idea.

Well, 26,000 registered. Many thousands actually built a business idea. And we're going to be— we have On stage with us as the judges there is going to be Palmer and Ben Lam and Mark Pincus and Logan Kilpatrick from Google. And I think the important thing for everyone in the audience— and the event is capped at 1,500 people. And we're being very selective on who's there. We're going to be analyzing how they did it. Our goal with Build with Gemini XPRIZE is teach people how to fish. Instead of waiting to go get a job, find a problem that you're passionate about solving and code it up and build a business. And it's, it's, you know, the goal is demonstrate anybody can do this. 26,000 people entered this competition to do that. It's gonna be great.

Unbelievable.

I'll mention one other thing. We have the Future of Vision XPRIZE as well, which is culminating on that day. We have 5,000, over 5,000 people who entered this, you know, largest world film competition. And Neil deGrasse Tyson and Neal Stephenson will be judges in that. You know, I'm pumped.

Amazing. Can't wait.

So our next story, Memory is the Bottleneck. I had a chance to meet with the leadership of SK Hynix and Solidigm. We'll talk about them in a moment. And I was so blown away by that meeting at how it's not GPUs It's actually memory is the rate limiter, right? So I posted this on X: memory, not compute, is the rate limiter for the agentic era. And Elon posted back saying, few realize this. And then, of course, my tweet, you know, exploded to 7,000 likes as a result of Elon's interaction, which I appreciate you, Elon, for doing that. And the story is, is What we're seeing is a situation where in the agentic era, where your agent wants to remember everything about you, we need to have more memory. So, you know, the first story here is the stratospheric increase in memory prices. They've climbed 500% in 12 months. Hyperscalers are reportedly locking in their global DRAM production rates for the next 12 months.

Wow.

through 2027. SK Hynix CEO warned that 2027 will be the worst year for memory supply industry's history and will, uh, you know, demand will outstrip production capacity well into the 2030s. The second story is that only 2% of the world's memory chips are made in the U.S. While the global production rises 20% annually, AI demand for memory is growing at a rate closer to 200%. The third story, finally, is that Elon's TeraFab will manufacture memory in-house alongside logic chips, which is the strategic decision made by TeraFab to go vertically across the entire AI manufacturing platform. Finally, Solidigm, SK Hynix's U.S.-based NAND and enterprise SSD business, has staged a dramatic turnaround according to the Nasdaq listing. First, its first-half revenues hit $8.6 billion, with net margin gains going from 3.9% to 47.7%. So the memory story is simple. AI needs memory to think. Every GPU needs 4 to 6 times its cost in memory to function. As models get larger and agentic context windows expand, memory demand is growing faster than compute demands. Alex?

Yeah, a few different aspects here. If you remember during the pandemic when there was a toilet paper shortage, part of the— I mean, this is like one of my mental models for one of the streams here. There's a toilet paper shortage in part because during the pandemic people stopped going to restaurants and to businesses. And so as a result, all of the toilet paper and, you know, various other artifacts that were designed for enterprise consumption were suddenly being rerouted to consumer, and that led to all sorts of supply chain hiccups. Similarly here, the shape of memory consumption by frontier models is pretty different than the shape of memory consumption by applications historically. Like 10, 20 years ago, if you were using, I don't know, Microsoft Word, the amount of memory that was actually needed, far lower. Whereas if you have like a trillion-dollar model where every layer, you know, transformer-type architecture where every layer for the purpose of forward propagation needs to be loaded into some form of memory in order to do matrix multiplies, that has a very, very different memory footprint than just say Microsoft Word from 20 years ago. So that creates enormous pressure both on the supply chain, open paren, the memory and storage industry has historically been boom-bust and clays Christensen and others have written about this, creating a sort of paranoia by those in the supply chain of when the next bust is going to come around, resulting in them being paranoid of overbuilding, resulting in them being unwilling to respond elastically to demand, resulting in these crazy price swings. Because if you're not building enough supply chain infra in the memory industry to meet this now enormous demand for memory, The prices go up because the supply isn't going up. Economics 101. Close paren. There is a second angle here, which is the shape, the physical shape of memory itself. If you look at how memory historically has been consumed by compute, now like 20 years ago, again, I'll pick on Microsoft Word. It was very much what one might call like a von Neumann-type architecture. Crisp separation between the memory and the compute, more or less the equivalent of a Turing machine type tape where, okay, so you can randomly access different parts of memory and then you can load and then you do some compute and then you store back. But there's basically a clean separation between the compute part, which is the head, and the memory part. The advent of transformers and then frontier models has completely turned the whole situation upside down. People for decades— I remember 20 years ago when there were entire DARPA programs devoted to looking for what a post-von Neumann architecture would look like. Well, we found it, and HBM, I would argue, is like the foothills.

High-bandwidth memory, right?

High-bandwidth memory, which is the most highly sought-after form of memory. It's basically like 3D architecture where you have multiple memory layers physically sitting on top of the compute in one package. This is, I think, the foothills of a post-Von Neumann architecture where the memory is starting to finally merge with the compute. The memory transistors are right now layered on top of the compute transistors, but they're going to merge and we'll finally get past the Turing tape and the Von Neumann architecture. And I think that combined with the paranoia in the memory industry for the next bust, whenever it'll come, I think those 2 create this perfect storm where you see memory prices skyrocketing 5x in a year.

Let me put a number on it. When I was meeting with the SK Hynix leadership, they said the need right now is for them to 4x their manufacturing capacity. And 2xing it would cost them $1.5 trillion. And historically, this boom-bust, they would never make that large of an investment because there was always a bust. Afterwards.

And they're paranoid. They're scared of not surviving the next supercycle.

Yeah, exactly. Exactly right. Dave, do you want to jump in?

Well, TSMC said the exact same thing with GPU manufacturing. They were paranoid that if they ramped up the fabs, you know, which are, you know, the fabs are $20 to $40 billion each. So if they ramped up production in assumption that Nvidia would want more and Apple would want more, they would inevitably be overbuilt. And of course that's wrong. AI scales to infinity, demand scales to infinity. But the other counter-pressure is that photonic computing and new physics are imminent. And so you're like, HBM is a Rube Goldberg mess. It's absolutely— it's the biggest joke in the world because it's random access memory, but you're streaming sequential files off of it.

Yeah.

It's so insanely stupid. So it's the most valuable thing in the world right now, But better designs are going to come very soon because AI can invent things so, so quickly.

Yeah.

So everybody's scared to overbuild or overinvest.

So yeah, I have the same thing. The bottlenecks don't stop exponentials, right? They just redirect around that. We'll have capital going into new models, and, uh, just innovation will go towards eliminating all of this.

A fun factoid that's floating around, just to Dave's point regarding the value, I think the latest statistic was HBM on a per-mass basis is worth approximately, literally, half its weight in gold. So, if this keeps up, forget about gold, forget about precious metals. Just— this is not investment advice— hoard HBM.

Well, actually, if you look at the chips before they go into the packages, because the packages are 99% of the way or 90% of the weight, they're massively more valuable than gold. It's actually, I think, the most valuable thing in the world that you can put in a shoebox and carry around is unpackaged memory chips.

It's crazy.

Wow. Emad, any opinion here?

Yeah, no, the memory right now is about a third of all the infrastructure spend, and next year it'll go to 50%. And the market finds a way. Like, this is ridiculous. So I think that it may be that we don't find a breakthrough, but I wouldn't bet against it. I think that the fact that you have this really complicated HBM storing static weights makes no sense whatsoever.

No sense.

And as model weights satisfy and standardize, especially for things like being a decent doctor or something like that for a medical set of weights, you'll move to etching, you'll move to these other things. And then workloads will migrate because you don't have to pay half of a data center build-out for literally memory.

At the same time, the frontier can still push it way further than we can This is literally exactly why we founded Quantum.ai, Q-U-A-N-T-U-M.ai, but also why Talus just got acquired. I don't know if you saw that in the news, but Talus, they're not moving the weights. They're etching them into silicon or into wire on the chip. And then they're massively more efficient because they're not moving around. So, it's a huge breakthrough. There are all kinds of problems with manufacturing because once you've etched the weights, then they're frozen. And if somebody retrains a better model, you want to be able to say, okay, now I need to swap to those new chips. And our whole supply chain isn't ready for that rapid of an iteration. But you're literally looking at 100 to 1,000x performance gain if you edge the weights. So, lots of opportunity coming in this area, which is only going to— that's on top of the 10,000x we were talking about, by the way.

And everybody, this demand should be obvious, right? You want your agents to remember everything about you. you, right? Every interaction, build a world model for you that understands you, and that takes memory. And the more agents, the more memory, and it's very rapidly outstripping the importance of GPUs.

All right, Peter, just to refine that point a little bit, there's something even more scandalous, which is I don't actually think at the end of the day individuals have that much information about them that's worth remembering. It's— but there's enormous mutual information shared between an individual It's actually the world knowledge that's what's worth remembering. If a model knows basically substantially everything about the world, then it knows most of the information about the individual as well. I would argue it's world knowledge that the model has to keep in memory and weights more than individual personalized knowledge.

When you standardize that, as we discussed earlier, these things are converging, then you can have a reasonable engine with world knowledge. That is Etched. And the other company that's been etching is Etched. That's the name of the company. It just hit $21 billion in valuation.

Did they really?

I had a chance to seed invest in that, and I missed it. There's Etched.

There's Architect Labs. We're talking my own book. There are a bunch of companies pursuing this.

All right. I'm going to move us into the world of robotics, give you guys an update on what's going on in the robot world. Unitree's newest humanoid robot, only 3 months in development, broke every human standing jump and speed record. Standing jump at 2 meters and a top speed of 12.66 meters per second, beating the human record set by Usain Bolt, who reached 12.4 meters per second during his 9.58-second 100-meter world record. Let's take a look at 2 videos here just for fun. And then another video of that superhuman race because it ends in a nice little scenario here. Oh my God, it needs brakes. Salim, I'm gonna go to you first on this.

Um, where do you want me to go? Okay, look, you, you— I think we should stop trying to make robots human, right? Like, just make them economically useful.

Salim, you don't disappoint. Like, well, I just—

why?

We are optimized for 4 billion years to survive and procreate, right? If you want a mining robot, make a mining robot. Give it wheels, give it whatever, give it multiple arms. By the way, I just want to just say thank you to all the fans that send me images of like 6-armed robots and stuff. It's awesome. Totally, totally love it. So really, really appreciate it. But I think the big story here is a 3-month compression loop here, right? The iteration cycle is shrinking dramatically, and there's multiple exponential curves happening. Like, you've got AI, you've got simulation, you've got batteries, you've got actuators all multiplying. And so this is going to be— you're getting hardware now to the same loop cycle as you have pretty much software, and that's huge.

Alex?

Yeah, I was studying how Unitree achieved Superman, their robot And it appears that what they did— this is based on publicly available information— was they shifted the mass budget for the humanoid robot around to optimise it for leg performance. So they subtracted mass from parts of the upper body that weren't slow.

They're benchmaxing.

They are leg benchmaxing. They're leg maxing. And so, so I, through the, through the lens of benchmaxing or leg maxing, This makes me think that maybe to take the counterpoint to Salim's comment about, oh, we want multiple body shapes, I actually think this is not a stable equilibrium. I do not think that we end up in a world where we have some robots that have like really strong legs but really weak upper bodies, and other robots that look totally non-human but have really strong arms or whatever. I think that would be an By analogy, if you remember, like in the 1980s, before the broad advent of personal computers, or the '70s, or call it '70s or early '80s, before we had broad general-purpose PCs and there were dedicated word processing devices and dedicated other devices, and we had Wang Computer in Massachusetts, I don't think that's the way of the future. I think my prediction is we will wind up with generally capable robots that are As with generalist models ultimately devouring and subsuming all of these specialist models, no, I don't think we're going to wind up with super strong-legged robots. They're going to be general purpose, and they're going to be general body plan, and they'll be good at everything, would be my bet.

Just give it wheels. Just put wheels on it.

Wheels aren't general purpose. We learned this from Doctor Who. The Daleks originally in the original— Imad, this is maybe your neck of the woods, right? The Daleks used to not be able to climb stairs, and then I guess in the new Doctor Who they can climb stairs. We want generally capable robots, and I think that means legs in the short term and maybe nanites in the long term.

Yes, nanites. We're back to Diamond Age. What makes this interesting is the human form, right? Because we have supersonic jets and we've got rockets that can go as fast and leap higher than anything else, But it's because we sort of anthropomorphize them that's interesting. And I think as we're moving in that direction, you know, I went to the Enhanced Games back 4 months ago, and I think we're going to start to optimize humans, and we're going to watch the robots do everything they can do and optimize humans to do what they can do. Dave, what are your thoughts here?

Well, as an investment theme post-AGI, post-ASI, which is very, very soon, robotics is Just fertile because, because of exactly what Celine's been saying for a long time. There's so many form factors and so many shapes and sizes and innovations. And, you know, the AI mechanical design is starting to work for real. You can just vibe up parts. And also the manufacturing supply chain, you know, is starting to get invested for the first time in, I guess, since Detroit. So 30, 40 years. I didn't know until Alvin said it, but the US, you know, had 50% of the world's manufacturing capacity back in the peak of of our manufacturing days. And now it's 1/3 China. And he said it was about 15% US. But it's starting to get huge amounts of investment. And the returns on that are going to be phenomenal. So that'll last a while. So that's great.

Imad, any thoughts to take us out on this story?

Yeah, I think that these types of robots will be banned from the streets.

So interesting.

It's like—

so I mean, the stuff is strong as fuck.

But you stop as strong as you are. They're going to hit the wall. That's not good.

I mean, it's obvious that they would be beyond human capability, right? But now they have coordination not to hit the wall, as it were. But you don't want to have superhuman robots on the street because you will have accidents, you'll have issues just like cars. But that opens up to soft robots, you know, like—

Imad, one second. You know, there's going to be a point at which they're running an AGI model and they can avoid Is it that they're not trustworthy? What would keep them off the streets?

No, the extreme robots which have beyond-human capabilities will be kept off the streets or they'll be regulated.

That's kind of my contention here.

Or they'll be military.

Well, yeah.

The 1X robots, for example, some of us will be getting ours. They're nice and soft, you know? They have all these things. They can't twist off someone's head or accidentally punch a hole in them. Whereas these now, you will have the extreme robots like the Ferraris, but most people will get I do agree with Imad, for what it's worth.

I think just like we see regulation in truck sizes versus car sizes versus motorcycle sizes and what can be supported on certain roads, or laser intensities and laser power, like 5 milliwatt above versus below regimes, I totally buy that in the near future we'll see, well, This road is zoned for the following, like, power density of robot, or this— this will probably see, like, classes of them. And certain— we'll see, like, consumer-grade robot classes versus industrial versus military-grade robot classes with different power densities or torque densities. Totally buy that.

Everybody, welcome to the health section of Moonshots, brought to you by Fountain Life. You know, AI is impacting every aspect of our lives— how we teach our kids, how we do our business, But one of the most important things that AI can deliver to us is health. And one of the things I think about when, you know, shooting for 100, 120 is, am I gonna have the cognitive health to be able to think clearly and keep my wits about me for the next 50 years? I'm joined here today by Dr. Dawn Musalem, the Chief Medical Officer of Fountain Life and a member of my Fountain Life medical team. Dawn, a pleasure. So Dawn, talk to me about brain health.

Brain health. You know, you're right. This is the number one concern people coming into Fountain Life have is, will I remember the name of my child and the face of my loved one? 45% of dementia cases are entirely preventable with lifestyle. And what was really intriguing to me, Peter, is that a quarter of our members had advanced brain age, but over 13 months of us really helping them live healthier lifestyles, eating healthier, moving their body regularly, and optimizing sleep, People overlook that so often, but that sleep optimization is critical for our brain health. What we showed is that we were able to improve the brain age in 46% of those individuals. That's a powerful number.

That's amazing. You know, one of the things I love about Fountain is we're constantly searching the world for the most advanced therapeutics and bringing them to our members. So for me and all of you, I hope that you appreciate the fact that you can become the CEO of your own health. You can make sure that you've got the cognitive clarity for the next 50 years. Come and check it out, fountainlife.com/peter, to learn more and become the CEO of your health. Now back to the episode. I'm excited about this next story. It's about a friend, Keller Clifton, the CEO of Zipline. I had Keller on stage at the Abundance Summit last year along with Dara from, from Uber. So this week, Zipline announced that they are scaling to provide Uber Eats with more than 1 million autonomous deliveries per day. Uber and Zipline formalized a partnership targeting a million autonomous drone deliveries carrying your Uber Eats to you. And I guarantee you, when that becomes available, I'm going to be using that all the time.

Every day.

Yeah, it's going to be fun. It's like, it's like entertainment while, while you get your food. Uh, and Uber is also doing a significant investment into Zipline. Keller's framing We have entered the scaling era for robotics and physical AI. Dave, you've been saying that 1 million deliveries per day is not a pilot program; it's infrastructure. Each delivery replaces a human driver, a car trip, and the associated emissions. At a million per day, Zipline is moving more packages than many national postal services. Amazing! Let's take a quick look at this video from. from Keller and from Dara, and then we'll chat about it.

We're super excited to have Dara here today. We're announcing a partnership between Uber and Zipline. That involves an investment, and more than that, a partnership for Zipline to power home delivery of hopefully a million and then more Uber Eats deliveries to your home incredibly quickly, incredibly delightfully. And that's a million deliveries a day.

Amazing.

Gentlemen, who wants to jump in first on this?

I'll maybe just comment. I think this is a clever and also inevitable move by Dara and more generally by Uber. Dara has taken Uber with a number of acquisitions, strategic acquisitions over the past few years in the direction of being a mobility aggregator. And thus far, Uber has, other than maybe like Uber air taxi type initiatives, has basically been focused on ground-based mobility. And I think this is a— this represents a serious move in the direction of aerial mobility. Of course, China has had this now for at least a couple of years with the ubiquity of air-based drone delivery of foodstuffs and other matters. I think this is a very positive move for the West. I think the elephant in the room, though, from Uber's perspective is if you think back to when Uber basically hollowed out Carnegie Mellon University's robotics department in order to try to build up its in-house robotics capabilities, and that was more or less a disaster and didn't quite work out, and there were lawsuits with Waymo and otherwise. I think this is— call this Uber's mobility plus autonomy 2.0 strategy, where Uber focuses on being An aggregator at the software layer.

A platform.

An aggregator in particular. Third parties, including, by the way, Waymo, are providing all of the physical autonomy, and Uber is just the demand aggregator for everyone to consume mobility from all of these different third-party providers. I think that works really well for Uber as long as it maintains competition, healthful competition among all of its its mobility suppliers. It's bad for Uber if the industry verticalizes and if Waymo or Zipline and someone else just decides, we don't need Uber as an aggregator, we'll just do an end run and sell directly to the customers.

You know what I think is incredibly cool, just incredibly cool, is if you drive down any street in America, any suburban street, any urban street, it goes fast food, car dealer, fast food, car dealer, fast food, car dealer. And in the very near future, the food will be off the main street and it'll just pop over the mountain and drop on your, you know, your lap, and the car dealer will— the car will drive to you.

There's no—

there's no—

That's so brilliant, Dave.

It's gonna be so nice.

Oh my God.

Yeah, it'll totally reshape things. This was very EXO, by the way, because you've got Uber, as we've said, aggregating demand, and then Zipline gives you all the autonomous, uh, assets. Infrastructure.

Amazing.

Yeah, by the way, but I think Alex's point is really important that if they try and kind of control it. But what we heard from Dara last year on stage was that he's planning on creating as many partnerships as possible.

Of course.

And becoming kind of like that wiring. And that, I think, is a smart play.

By the way, if people are interested in the Abundance Summit, it's in March every year. We bring in, you know, CEOs like Keller Clifton and Dara from Uber and Elon and across all of these areas. It's a March 7th through 12th next year. You can go to abundance360.com. The mates are going to be there as well. Uh, you know, yeah, go ahead.

I'll make a— I'll make one forward-looking prediction here, right? Because this is something we probably could have seen coming. Let me— let's bridge forward. Imagine if they now do a partnership with Shopify and every small merchant gets an Amazon-grade logistics capability. That will change everything.

Brilliant.

Yeah, I'll invert your prediction, Salim, because Amazon obviously has their own in-house drone delivery capability, which has been delayed for like 3 years. It's crazy for regulatory reasons, is my understanding, not because there's something technically wrong about it. Just like this is new for the West, at least. Do you think Zipline ends up being a highly appetizing acquisition target for, say, a Shopify to in-house its delivery capabilities against Amazon.

Interesting. Yes, yes, yes is the answer.

Yes, I think so too.

Yeah, great thought, great thought.

And then this is, by the way, I should add, like, the hot news for the past day is the video going around social media of a woman watching in horror as one of these drones delivers her package into her swimming pool. And it sinks.

Well, you know, we just had back to back on this pod. Etched and Zipline are both companies where on founding day you're like, really? Can that— there's no way. The amount of moving parts required for that to work. And now you're looking at, by the way, $20 billion and whatever.

We're going to have Keller on this pod. Uh, Keller's agreed to come on the pod, uh, and talk to us about this. Maybe we'll do it live over at Zipline. It's up to you guys. What you want to do. And then, by the way—

yeah, sorry, go ahead.

No, no. And then we, we also have, um, uh, you know, a lot of incredible guests that are coming. We're going to bring back our, our dear friend, the CEO of Figure AI. Uh, Brett's coming back on the show, or we're going out to him. So that's gonna be fun. You're gonna say, Salim?

No, I just want to say this whole, um, delivery by drone, we had a Singularity University project in 2010 that did this, right? And they looked at Africa and they realized that Africa leapfrogged the entire landline and went to a billion mobile handsets. Why would you spend a trillion dollars putting roads across Africa? Just go straight to drone delivery. And they demoed that, and that apparently inspired Amazon, and I think cascading down a lot of the others here.

And the backstory here is that Keller Clifton, a San Francisco-based company, began operations in Africa because they were able to take care— take advantage of regulatory arbitrage. You know, the country wanted them there, and they developed operations and safety and then came back to the US.

Yeah, what Rwanda did, uh, where a lot of the starters was, they basically said there's a 3-dimensional tube across the country, like a superhighway If you keep your drone in that 3-dimensional corridor, you can do whatever you want. And that allowed people to really go and play with things. Really, really big breakthrough.

Yeah.

All right, I'm gonna move us to our final segment on health, a really important one for everybody. Health is your new wealth. So 3 exciting stories. The first story, perhaps the most significant, is out of Moderna and Merck announcing that their mRNA cancer vaccine Succeeded in a late-stage melanoma trial, marking the first Phase 3 validation of personalized mRNA immunotherapy. So more than 8,500 people in the United States are expected to die from this deadly skin cancer this year alone. This is the cutting edge of science, right? The vaccine works by sequencing a patient's tumor mutations, identifying neoantigens, unique antigens for that cancer, and then manufacturing a custom mRNA vaccine that trains the patient's immune system to attack the tumor. So let me unpack this a little bit. So the first thing you're doing is you do a surgical resection of that tumor. You grab tissue, you do a whole exome and RNA sequencing, you feed that into a machine learning model that's looking for unique antigens. It ranks them like, here's the most unique surface antigen. mRNA encoding up to 34 patient-specific antigen targets, and then it's manufactured and shipped in only 8 weeks. Every mRNA vaccine developed using machine learning creates a unique mRNA sequence for every patient, so vaccine for you is not the same as the vaccine for me. The Phase 2 results stopped recurrence of death by 49%, and distant metastasis or death by 59% over 5 years, and it's expected that the cost of this treatment will be as low as $5,000. Moderna stock surged 110% after announcing these results. Alex, I'm gonna go to you first on this.

So many thoughts on this. So the superficial thought— this is obviously a great day for cancer survivors and for treating cancer in general. That's the superficial thought. I'll go a level deeper. First of all, I want to browbeat Moderna and Merck just a little bit for naming this drug. So the drug's name, the official name is— I'll see if I get this right— istismaran, I think is how it's pronounced. So in doing research, turns out istismar in Turkish is the word for exploitation or abuse. So just pro tip to Moderna and Merck, please, before the rest of the world figures out what the name of this drug is, is rename it from Istismaran to something that works well in every time zone.

Super Duper Moran.

Yeah, seriously.

But these are FDA-approved names. It's crazy how they name this stuff.

They don't care about Turkey, I guess.

But they don't care about, you know, having a neologism roll off your tongue onto the floor either.

That's right. But on a more serious note, so I remember the National Nanotechnology Initiative in the early 2000s. when Eric Drexler et al. sold the U.S. Congress on spending billions of dollars on nanotech, going back to Diamond Age, on this thesis that we would have nanorobots going through the human bloodstream zapping cancer cells. Well, guess what? It's 2026, and we caught up with the future. They're not diamondoid nanorobots. They're lipid nanoparticles with mRNA snippets, 34 different mRNA sequences. And so they're like soft robots. They're not like this machine phase Drexlerian nanorobots, but nonetheless, these are primitive nanorobots that for the first time— this is the first successful phase 3 success for an mRNA cancer vaccine. This is the first, but not the last. There are going to be so many of these. All you need to do is look at Moderna's pipeline, which I think they do an extraordinary job. And they're only a few blocks away from me here in Cambridge, of maintaining a public pipeline website where you could see the clinical stage of every one of their vaccines for infectious disease, for some rare diseases, for cancers, for other classes of diseases. This is a general-purpose platform. This is arguably what we wanted 20 years ago out of nanorobots. It's just that they're soft and they're made of fat. They're not made out of— They're not made out of metal.

Yeah.

Hard diamond stuff. And so that's one point. Second point I just want to highlight, there's a technology underneath this that I think is going wildly underpublicized, which is the RNA sequencing technology that's enabling this to be personalized. So Peter, you touched on the first half of this, which is the RNA sequencing and mRNA sequencing of the tumor, but in order to calibrate what the right expression profile is for the tumor, you also need a second mRNA sequencing profile from the bloodstream to know what's abnormally expressed in the tumor and what isn't. So that's from another company called Personalis that has what they call their Next Personal Sequencing technology that they originally developed, in my understanding, to do blood-based trace cancer detection. So if asked the question, how does this all look?

Like GRAIL, it's liquid biopsy. Yeah.

So I think, like, extrapolate out a few years, maybe we won't even need— for personalized cancer therapy, maybe we won't even need to sequence the tumor itself. Maybe we'll just get all of this from the bloodstream and be able to do continuous medical monitoring via these models.

Yeah, yeah, fully agree. And 2 things I was gonna say, a quick congratulations to a friend, Stéphane Bancel, who's the CEO of Moderna. You know, they got a lot of negative news on the COVID vaccines. I mean, even though they came out with the vaccine very rapidly, the work that Moderna is doing is amazing on personalized cancer vaccines. They're also building out the ability, you know, there's a lot of endemic, you know, CMV and Epstein-Barr virus out there, you know, in the world population. And they're building out the ability for you to actually fight those infections internally to yourself. So a lot of headroom for Moderna here as they dive in across the board and use this technology. Salim?

2 things that struck out to me. One is the regulatory structure that's allowing for personalization. That's a huge thing. We've never been able to do that before. So that opens up the floodgates for all sorts of things. Daniel Kraft talks a lot about we're getting into personalized medicine, and the fact that we can regulatory navigate How do you deal with the sample size of n of 1, right? And the second thing that struck out to me was Raymond McCauley, who years ago said these mRNA vaccines, I think he would put it, it's the first battle in the last war against all disease. And you're like, wow. So that, I'd love to see this fruition coming to play.

Dave or Imad, do you want to hop in?

Well, this one strikes close to home for me because my daughter works at Moderna. and she's my go-to on— but she's been telling me and sending me research reports for months. This is not a secret. You know, the stock went up yesterday. It almost tripled yesterday, the biggest one-day pop in any S&P 500 company of all time.

Wow.

By a wide margin. Just a massive— like, and so one thing immediately came to mind is, well, should have bought the stock. Listen to your own daughter. That's advice number one. But number two is, Because back, remember when Enron and Tyco had all those fraud issues? They passed the Sarbanes-Oxley Act and a bunch of other laws. One of the byproducts of those laws is that a Wall Street analyst can't trade the stocks that they cover. And so, everyone who I know who was in that job is like, well, why would I study Moderna or other super high-tech stuff, learn all about it, and then not be able to trade? So, they all quit. And the byproduct of that is that, you know, stock market is now dominated by tech, which is very complicated to understand. But the research community is the worst I've ever seen in, in Wall Street history. And then not only that, the indexes have taken over half the market. They don't think at all. And so the amount of useful information is at an all-time low when the things that need to be explained and understood are at an all-time high. But it was no secret it, that this Moderna platform can basically be used for any form of cancer and that it's highly likely to work. It's just a question of time. The research is all out there. This wasn't like some kind of insider surprise. Anyone, any good analyst studying this would've seen this coming.

Iman?

Yeah.

I mean, I think this is the interesting thing. I don't think any of us are surprised by this result.

Right.

And we won't be surprised when other ones go. But our current regulatory regime means they'll have to go through the same process over and over and over again. When really, screw cancer. Let's actually think about this from first principles. When we have systems like this that are very targeted and upgrade the regulations so this can actually get out to people faster to save their lives.

Well, our next story is going to take us there, right? Because if you can simulate all of this in silico and actually prove that it works, we should be able to do the studies in, you know, a GPU cluster and say, yep, it's safe, let it go. So let me turn to that story. And it's one I've been excited about and tracking.

I know.

Alex, you as well. So our final story here is about AidoCell. A-I-D-O is how it's spelled. A general-purpose cell simulator that maintains cellular state, accepts interventions, and predicts multimodal biological outcomes. The goal: make experiments computable before they're run in the lab. You know, cell simulation could reduce wet lab experiments 1,000-fold. If AidoCell can predict which experiments will work instead of testing 10,000 compounds in a wet lab, you can simulate them digitally, again, in silico, and test only the top 10 that the simulator says is going to work. This is going to drop the cost by orders of magnitude. Let's watch a quick video here, and then we'll go to the conversation.

Traditionally, biologists have relied on lab experiments, in vitro models, to understand how cells behave. Now they can use ItoCell, GenBio AI's virtual cell world model, model to simulate the same biology in silico, combining multimodal, multiscale detail with sequential experimentation in ways no microscope or wet lab assay could achieve alone. At its core, the IDO platform is a rich, stateful simulation environment powered by the first-world model of a human cell, one that predicts what happens at every level and remembers every change you make along the way. From DNA and RNA to protein interactions, structures, and localization to the whole cell, including cell-painted morphology. The model simulates cellular responses to genetic perturbations and treatments with small or large molecules. These can be layered in a sequence, providing the ability to watch the cell's full multiomic response with each step. The kind of insight that could mean computationally testing new drugs designed in cellular context Before they ever reach the lab bench. AIDO can be easily adapted to new cell types and indications. Using your own data, you can build models tailored to your research questions to simulate different cell types from shape down to genes and their protein structures, where they end up in the cell, what they interact with, and how they affect cellular responses.

Wow. This is the foothills of longevity escape velocity. I've been waiting for this forever.

Alex? Medicine is cooked. This is what— if people like the catchphrases, read my lips, medicine is cooked. This is what the end of medicine looks like. It looks like a virtual cell.

Or the beginning of longevity escape velocity, let's put it in a positive sense.

I actually think we can get probably to LEV without solving all of medicine. My bet is it'll be probably a class of molecules, maybe like 4th or 5th generation GLP-1s that get us to LEV. I think this is actually a superset of getting us to longevity escape velocity. I think this is like— this is how we cure all disease everywhere. And the way we do it is we build a virtual cell. It's just like we didn't actually have to solve human intelligence to solve AGI. It turns out you can just get AGI from compressing general human knowledge. It's not that hard in principle. Similarly, I think this is how we solve all disease. It's not that hard. In principle, you simply train the world's best foundation model to model all cell states and all interventions against cells, and then you do like an AlphaGo-type tree search against possible interventions to discover how to steer a virtual cell state from a diseased state to a healthy state, and then generalize that to tissue and organisms, and boom, you've solved all human disease. I think that's like the end— This is hyperbole.

This is hyper-personalized for you, right? You insert your DNA sequence, right, and your current blood chemistries and all of that, and there's an in silico model of your biology, and it will tell you whether this drug works for you or doesn't. Celine?

Yes, but I also, just on that, I don't want to like over-romanticize the personalized aspect.

It's—

an ideal virtual cell is as personalized for you as, say, if you feed a quote-unquote personalized prompt to ChatGPT, the output is personalized for you. Well, yeah, superficially it's a function of the inputs, but actually it's a generalist model.

Salim, your thoughts, pal?

Well, this has been a trend. I'll go back to the biotech stuff, right? We've been, we've been turning biology into information. When you turn something into information, it hops on the exponential curve, and we're seeing this go through live And the, you know, each of us have what, 50 trillion cells in the human body?

Yep.

Right, roughly. Essentially, when you can model that, essentially a human being becomes a software engineering problem. And we have really good techniques to navigate software, read, write, understand, etc., etc. And the phenomenal, amazing thing for me, we've done a good job in reading. You know, you have reading, writing, Reading comprehension, right? When you're trying to learn a new language, in this case the language of biology. We've done a pretty good job of reading. We've started to do writing with CRISPR and now these mRNA vaccines, etc. This gives us a huge depth into the comprehension side with the digital twinning that can take place. So holy crap, this opens up the door. Uh, I would go with Alex's comment that, uh, medical is cooked.

Imad?

Yeah, like, this is kind of my hope for what the Genesis Project would be like. Um, it seems very straightforward to me now that if we had a Manhattan Project to cure disease through in silico massive, well, human body models, cell models, and organizing all our collective knowledge on cancer and autism and all these other things, we will definitely get a result. Like, I'm not even like—

It's not going to be a government program. The labs are going to do that for us.

But I'm like, why don't we actually get together and get governments to put into a Manhattan Project type thing and just have a straight shot at it and make all the data open? This could be the biggest—

I think your point is well taken. At least the government is sitting on a lot of data, and the government could externalise all the data to the private labs like CZI and IDO Cell and others who are all building foundation models, make it a public good that they can Yeah, we've done that in the UK where you have access to this.

Every government should follow suit. And this fits with what we talked about earlier with Anthropic. Again, what are they going to apply their computation to? They will build a human cell model. They will build a whole body model. But I'd prefer for those to be public goods and us to say, let's cure cancer. Let's address all these negative kind of things and let's understand the body like never before. Again, that's much better than building an atom bomb even. because it will have the biggest impact on humanity ever.

Dave?

Well, just for anyone listening who's not a biotechnologist and they're like, well, I'm not gonna build a full cell simulator, I have no idea how to do that, you're thinking about it the wrong way. You heard earlier on the pod that we're looking at 10,000x expansion in AI. With some other innovations, it could be more like a million x. It's totally data-starved. The full cell simulator is a way that it can design thousands or hundreds of thousands of experiments and get reasonably good test results back through a simulation rather than having to run millions of assays. That also applies in all kinds of other areas where every investment we've made in a company like Merckor or Mikado that is wrestling with new types of data to feed the AI, they're thriving and growing and making money and valuating. Merckor's worth $40 billion or $20 now, $40 by the end of the year. They're just absolutely killing it. But every field of endeavor is going to be data-starved. So no matter what you know, you probably know a field that needs to supply data back to the great AI. The Full-Cell Simulator is just the perfect biotech solution, but every industry has a solution.

Yeah.

And so competitive too. I mean, it's probably worth noting that this is from a company co-founded by David Baker, who shared the 2024 Nobel Prize in Chemistry with DEMIS for solving protein folding. This is the next big thing, next grand challenge arguably in biology/medicine after— now that structural biology arguably has been solved. Solve whole cell simulation, and then you're halfway to solving all disease.

Oh, love it. All right, to call out to our creatives out there, please send us your outro music videos. We're at a paucity. Send them to media@diamandis.com. We love your outro videos. You're going to want to see it. I'll pull it before we go to our AMA. A quick note again, follow us on X @Moonshots_pod. We're going to be putting out clips and putting out these podcast recordings on X as well. And join us at the Moonshot Summit. Go to moonshots.com to apply. Gentlemen, Salim and I have an AMA with the Abundance community in 8 minutes. So I'm going to suggest we speedrun the AMA if you don't mind.

Sure, we can do the AMA in 8 minutes.

We're going to do the answers. Yes. Okay.

Warm up.

All right, Imad, you pick one first.

Can we keep making Frontier models more energy efficient instead of building all this new power? @Brion75. Yep. I mean, that's— necessity is the mother of innovation. As we run out of energy, as we run out of RAM, you're going to optimize immensely, and it'll be a big boon for everyone.

Salim.

Will we see an XPRIZE aimed at solving the electricity supply problem? We actually proposed in the last Visionary or a couple of Visionaries ago enough energy storage off-grid to keep a village or a town energy sufficient for 3 days, but it looks like the market will take care of that.

Yeah.

And regulatory is the problem, so it doesn't really serve as an XPRIZE where you need huge technology This serves better as a— this is a regulatory issue and a market issue.

Dave, over to you. I'd like 4.

If China has the advantage on power and the US has the advantage on chips, who actually has the real AI advantage? Definitely chips are, you know, power is a problem, but we need 100 gigawatts by the end of the decade. We already manufacture a terawatt in the US. So 10% of power will go to AI by the end of the decade. We'll get that far. Then we'll be really desperate for more power. But between here and there, it's all about chips. Every chip, you know, that's why memory is up 5x. So that's the bigger advantage in the short run.

Yeah. Alex, number 2 is for you.

Number 2 asks, could interconnected microgrids popping up everywhere reduce the load on the main grid enough to matter? I think I would invert the question, invert the premise of the question. It's not that the load on the main grid is going to It's the exact opposite, that there's so much economic demand for the compute and the compute needs so much energy. Before long, unless we fully externalize all of the compute to orbit and the Dyson swarm, these data centers are going to be generating a surplus of energy that can be pushed back onto the grid and driving utility prices negative.

Yes. I mean, I want to, I want to make that point. You know, if you're arguing against a data center in your backyard, You're arguing against lower-cost energy and economic advantages for your community. Please understand that. All right, let's go— let's start with you, Alex, on this one.

Okay, so I'll take question number 8. At the current rate of improvement, how long until AI is more efficient per watt than the human brain? I think we're probably already there. So that's a hot take on this one. A lot of people have this maybe like fetishization of the Landauer limit, thinking, oh, well, we must really be far away in efficiency per watt from what biology is able to accomplish. Biology is actually wildly inefficient. Our whole organism and mammals in general were never optimized for compute, whereas silicon and CMOS and whatever comes after CMOS, maybe it'll be photonics, which I know Dave is of interest. It was optimized from scratch, was designed for compute. I don't think the human brain is as efficient as many think, and the argument can be made that actually if you look at a watts per task or watt-hours per task basis, the leading-edge GPUs may actually be already more efficient than human brains.

Especially if you take into account the amount of energy required to train up a human over the course of 20 years. Right.

Yeah. Lifetime total cost of ownership, as it were.

I'll put a pin in a corollary to that too. People use the power difference as a way to explain that AI thinking is very, very different from human thinking. I think that people will soon realize that it's not that different at all. The power difference will go away very quickly. But also this like, yeah, that's why it's nothing like us, will also gradually go away.

Salim.

I will take number 5. What's actually driving the rush? Why not keep energy growth at past levels and accept slower growth?

Yeah, because it hurts a little bit.

From WHDM DRN. I think the easy thing here is intelligence is looking more and more like a general-purpose input that will drive economic growth, right? And so having— saying let's have less intelligence, saying let's have less electricity or less internet. It's— you want more of it. And it improves research, improves drug discoveries. We saw all today, logistics, everything. So you want as much of it as possible. There's also competitive pressure. If one company slows down or country, then others won't. So it's not about accepting that capacity. You got to get your head around the abundance idea. All the thing we should be doing is accelerating energy abundance.

All right, Dave.

I'll take the easy one. Number 7, whatever happened to fuel cells? Actually, you know, Elon Musk started— his passion in life was ultracapacitors originally.

That was his Stanford thesis before he dropped out, before starting.

Well, whatever happened to those too? What happened is lithium batteries worked far, far better than anyone ever would have predicted, and they're still improving. So it sucked all the capital out of the other ideas. So that's all that happened.

All right, Emad, close us out here.

Yeah, if 71% of Americans oppose data centers, is this grid buildout actually going to help regular people's electricity or just make it scarcer and more expensive? David Harmon, Note 03. I mean, it's what you just said, Peter, right? It's going to make your electricity cheaper. More power is good. These things are not polluting. We need more data centers. We need more power, and we need to make sure it's all built right.

Amazing. And you guys did it. You did it in 8 minutes. Uh, and Salim, we've got a whole 60 seconds to get over to our Abundance community.

I'm going to be 2 minutes late. I got to get like some I'll get a little bit of food in me before we go to the next thing.

I'll cover for you in the meantime. All right, gentlemen.

Jesus Christ.

I love you all so much. This was such a fun pod today.

This was awesome.

Just so much. And your brilliance, you guys are amazing. So, so proud to have you as our Moonshot mates. Alex, Dave, Emad, Salim, thank you always. Thank you to our listeners. We love having you. And hopefully you find this, you know, a way of keeping up with what's going on in the world, 'cause we are in an accelerating singularity. And no time to sleep, no time to blink.

Don't take off the day off.

Don't get fatigued like I did.

Yeah. Take care, guys. Be well.

All right.

Thanks, Peter.