为什么美国政府与 NVIDIA 刚刚救了 Intel（以及这对 AI 意味着什么）| EP #195

The US cannot let Intel fail. It’s our one and only chip fab company that’s truly domestic. TSMC, Taiwan Semiconductor Manufacturing Company—“T” for Taiwan—has 66% market share in the chips that drive AI. China is only 90 miles away from Taiwan and thinks it’s part of its country.

This does remind me a bit of the 1997 investment by Microsoft. I would love to see a success story in the style of what happened with Apple after the 1997 Microsoft investment.

It’s almost identical. Everyone had lost faith in Apple; it was almost bankrupt. Everyone forgets that. Nobody wanted to buy it, and Bill Gates decided to make an investment. It reestablished Apple’s credibility and gave it the cash it needed to turn things around. Then Apple went on to become the most valuable company in the world.

The US cannot let Intel fail. It’s our one and only chip fab company that’s truly domestic. TSMC, Taiwan Semiconductor Manufacturing Company—“T” for Taiwan—has 66% market share in the chips that drive AI. China is only 90 miles away from Taiwan and thinks it’s part of its country, so there was no way the US government was going to let Intel fail.

Where would you invest your capital, knowing what you know about the massive growth we’re seeing? Who wants to go first? Now that’s a moonshot, ladies and gentlemen.

We got on text this morning and said, “Oh my God, there’s so much news going on. We just need to spin up a WTF episode because there’s breaking news that we want to share with our subscriber base.” Dave, I want to say thank you to everybody here for rearranging their day so we could have this conversation. It was pretty extraordinary, with Gianluca, Nick, and Dana getting all the slides together. I feel like we’re going to be doing an episode of this every day at the speed at which things are happening.

Hey, Peter, it’s my absolute favorite thing in the world to do. I postponed meetings with JPMorgan and a couple of public-company CEOs to do this, so it’s a pure win for me. Things I wasn’t eager to do have been moved off, and the things I’m loving doing are right in front of me.

I was talking to Alex a few days ago, and I said, “How do you spend your time?” He goes, “Preparing for the pod.”

So sad, sad but true. It’s a black hole in a wonderful way.

Today in particular is a really big news day, so we just have to do it. We’ll cover it.

Again, as I say to everybody listening and watching, this is a chance for me to upgrade my IQ points by 20. I hope you do as well. The only way to stay on top of this exponential growth—we are such local and linear thinkers—is to constantly upgrade or update your belief about what’s possible and where things are going. We think of our episodes here as the news that’s worth listening to. If you want to trade the Crisis News Network at night for an hour with us, we welcome you to.

All right, let’s jump in. The news here, Dave, is that you texted me this morning and said, “Did you see what’s happening with Intel and Nvidia?” That spun up the watershed moment of, “Okay, time to record again.” So, buddy, why don’t you kick us off here?

Rearrange your whole day. This is huge news. If you go back to our August 13 podcast, Tech Experts Break Down the Incoming AI-Crypto Collision, which we did with Eric Pulier, and go to minute 56, you’ll see us calling this out and predicting it. That included the fact that Leopold Aschenbrenner, the genius we’ve been closely following through his hedge fund, had a half-billion-dollar position in options he had put on Intel. Our best simulation today is that he’s up about 200–300% on those options today alone, as am I, by the way.

It’s a big day for me, a big day for him, and a big day for us following these brainiacs. That’s why Alex is here. The world is changing so quickly, and opportunities are changing so quickly.

What happened here? Exactly as we had thought, the US government invested in and bought 10% of Intel. Nvidia has followed suit, and I’m sure there’s some White House pressure behind this deal. Nvidia needs chip manufacturing. It’s essentially sold out for the indefinite future, and the constraint on all chip creation is the fabs. Intel has fabs, and they’re actually starting to work really well, too.

The stock is up 25–30% today alone based on this news. More importantly, it gives Intel the capital to finish building out its 18-angstrom process, which they call 18A. That’s an absolutely cutting-edge, front-of-the-queue kind of process, and it takes the risk of complete failure off the table for Intel.

I can’t believe we’re talking about 18-angstrom processes. We’re getting down to atomic-level precision here. It’s insane.

It’s nanotechnology.

It’s nanotech. It’s fun to talk about, isn’t it? It’s one of the many things we learned at MIT that were impossible, including anything past a gigahertz in clock speed. There’s a whole litany of them. These barriers just keep falling one at a time.

Yes.

As Richard Feynman said, there’s plenty of room at the bottom. So, Peter, if you’re a little anxious at night, worrying that we’re going to run out of nanometers or angstroms, there’s always picometers and femtometers after that.

I love Feynman’s paper, “There’s Plenty of Room at the Bottom.” I think it was 1957 or 1958 when he wrote what was essentially the first treatise that became the thought experiment around nanotechnology. Do you remember the prize that he put up at that time, Alex?

I think maybe you’re gesturing at the Feynman Prize hosted by the Foresight Institute.

Yes.

Well, no, no, no. This wasn’t that. This was back in the original—1958, I think. He basically said, “I will award a prize”—I’m going to guess at the numbers here, like $1,000—“for the first team that can create a working motor within a very small cubic dimension.” He expected it to drive nanotechnology, but he didn’t expect it to be won so quickly.

Very quickly, a graduate student came to him with a microscope, and on the microscope objective was this very small motor that he had created with fine tweezers and fine tools to win the Feynman Prize. That was not the intention, which is why you have to write the rules correctly.

But back to this one, Alex: How do you think about this move? History does rhyme. At the corporate level, this does remind me a bit of the 1997 investment by Microsoft. Microsoft, in exchange for an investment—Apple obviously wasn’t doing well at the time—made a deal with Apple.

This image shows Steve Jobs at the podium in his classic look and stature. On a giant screen is Bill Gates, a very young Bill Gates. What are Steve and Bill talking about here, Alex?

They were announcing to the audience what I think was a very surprising partnership. Microsoft Internet Explorer would be bundled with every new version of Mac OS as the default internet browser. There were several other parts of the agreement, including Microsoft Office being refreshed or renewed for Apple devices. But this was also, in some sense, a turning point for Apple.

I have a long history myself with Intel. In high school, I did the Intel International Science and Engineering Fair. I was a winner of the ISEF. I was a winner of the Intel Science Talent Search. I was a winner of the Intel Undergraduate Research Award. I have a lot of history and amazing feelings for Intel, and I would love to see a success story in the style of what happened with Apple after the 1997 Microsoft investment.

Salim, are you jealous here?

No. I’m always in awe of the intellectual fortress we have with us, so I love it.

This analogy was actually brilliant. It’s exactly 10%. It’s almost identical. You have a company that everyone has lost faith in. Apple was almost bankrupt; everyone forgets that. Nobody wanted to buy it, and Bill Gates decided to make an investment. I think it was $150 million—very similar—but it was 10% of the company.

It reestablished Apple’s credibility and gave it the cash it needed to turn things around. Then Apple went on to become the most valuable company in the world. It’s really cool to dig that out of the archives because the deal structure and the whole dynamic are virtually identical to what you’re seeing today with Intel and Nvidia.

This is also a challenge for the audience: If you want to go back and pull up historical stock data, look at Apple’s stock prior to that investment by Microsoft and then compare it to Intel’s stock prior to this investment by Nvidia. I think you’ll see some incredible parallels. This is related.

This is me.

Yeah. What are we seeing here in this image, for those listening and not watching?

In the image, I'm teaching a class at MIT, Foundations of AI Ventures. Lex Fridman, who guest-lectured that day, is in the front row, which is really cool. One of the things I'm telling the class is that NVIDIA is going to become an AI company and not a graphics-chip company.

This is back in spring semester 2023. NVIDIA's stock at that point, according to this chart, was worth $500 billion. So it's now, what, $4 trillion to $4.5 trillion—up maybe 8× since that day. On that particular day, NVIDIA still had the majority of its revenue from graphics chips going into Xboxes and things like that, and not from AI. But this is an AI class, so I'm telling everybody in the class that AI is going to be so much bigger of a use case for NVIDIA.

I showed the stock ticker to the class, and the best thing about this picture is that kid in the front row—that student in the front row—who happens to be wearing a suit, which is really unusual, is on his iPhone. I'm guessing that he's buying NVIDIA stock during the lecture, so I'm hoping he is, actually.

I have a question for you guys. We've talked about this in the past: you're not going to give investment advice, and this is not investment advice. But if you were going to pick a stock or a field to invest in right now, you can tell me, “Listen, I'm going to continue to double down on NVIDIA and Intel.” That's fine; we've seen some incredible growth there. Where would you invest your capital, knowing what you know about the massive growth we're seeing? Who wants to go first?

If I only have a very small amount of money, I'm going after seed-stage startups—very smart people building AI companies that do something useful for the world, something that changes things. There are so many use cases, tens of thousands of untouched use cases, so you can't go wrong with those startup teams if they're smart people.

Then, if I have a lot of money and I'm trying to put it to work, it's the data-center buildout.

This whole everything we're talking about here with Intel—I love Alex's quote on this: the tiling of the entire Earth with compute, which is inevitable. There's no limit to the appetite of AI. There's no constraint. It will just keep improving humanity as you build more of it, but it requires massive amounts of data centers. That includes real estate and power.

But if you're going for public companies, which public companies are building out data centers today?

Well, CoreWeave—Crusoe is not public yet, but it will be. CoreWeave was the one you wanted, obviously. You'll see that on the next slide, actually, and there will be many more. We have a couple that are here in the incubator that hopefully will become publicly investable within a year or 2.

You saw with CoreWeave, when it came out, it was overlooked for maybe a month, and you had a great opportunity to get in. Leopold got in, and then it skyrocketed as people realized the demand is basically infinite.

Nice. Salim, do you have any favorites that you would—

Yeah. The obvious ones would be, say, Microsoft or NVIDIA, but I think that just plays to the narrative. I would go with NextEra Energy, which is exponential clean energy.

If we're tiling the world with chips, we need alternative energy to power those chips in a decentralized way. They're the biggest renewable utility, with decent dividends, et cetera.

There are 2 inflection points I want to point out here. In 2016, if you were building an energy-generation facility, it became cheaper to build a solar farm than to build a fossil-fuel plant. That was almost 10 years ago.

Amazing stat. So true.

But in 2019, we hit a second inflection point that's even more important: since 2019, it's been cheaper to build and operate a solar farm than just to operate a fossil-fuel plant. The capital and operating costs of solar are now cheaper than the operating costs of fossil fuels.

So now we know that government policy will naturally deprecate fossil fuels over time. All of that renewable energy is perfect for data-center power, et cetera. Low-cost power generated in all sorts of weird and wonderful places is going to be the future. So that's where I'd put a bit of a long-term play.

Yeah. How about you, Alex? Any bets?

The problem is, I think this is a trick question. If we're drinking our own Kool-Aid and we believe any of our forecasts about the imminence of superintelligence, recognize that most of the volume on U.S. equity markets is algorithmic in nature. The markets are already dominated by AI. AI knows substantially all of what we're discussing here. The market algos, the trading algos, have already priced in all of the headlines that we're discussing here.

So, really, for me to indicate any sort of preference is implicitly an assertion that I know better than the collective intelligence of all human traders and all AI traders, which in some sense would be contradictory to the underlying thesis that we're on the verge of superintelligence. So I choose—

That's a lovely and elegant copout, but I still call it a copout.

I choose the market overall—the superintelligence that it is.

Amazing.

If I'm thinking about big companies right now, I'm still believing that Google is the dominant player in this. While it's been priced in, knowing what Google has coming, I think it's going to dominate. At the same time, I think Elon, with xAI, is going to at least claim AGI first. We'll see how that moves things, but X is not a public company yet, whereas Tesla is.

I believe—basically, we'll talk about this in a little bit—he said, for Tesla, Optimus is 80% of its future value. I think the interesting trade would have been to have gotten back into Tesla when Elon dropped out of the government and started focusing once again.

I have a Google question, though, and this is for the 3 of you, which isn't easy to get a good answer to out in the world: why is Google not selling the TPU chips? They would make so much money if they put those—

Oh, yeah. No, I totally know the answer to that. The rate at which Gemini is growing, and Veo 3 is growing, means they can't get enough fab capacity to keep up with the demand.

So, completely consuming the entire supply.

Yeah.

I also think there's an interesting play to be made in uranium.

We're going to be heading toward a fission future. I'd love to go solar, but I don't think we're going to be building solar rapidly enough. If I look at the fusion timelines today, Helion is coming on with a small fusion plant in 2028 or 2029. Commonwealth Fusion is coming on in 2031.

We have a real problem in the United States. Eric Schmidt called this: “We're not chip-limited. We're not intelligence-limited. We're electron-limited.” We need to build SMRs and Generation IV nuclear reactors as quickly as we can, and we need uranium for that. So I might be making a bet on uranium-related ETFs.

Alex, do you not completely disagree with the premise?

Two points. First, I completely disagree with the premise of particular bets. Again, commodities are relatively efficient markets as well. If you subscribe to any variant or approximation of the EMH, the question doesn't even make sense.

Dave, you were going to say something before Alex was able to tell us how silly we were in such a beautiful and gentle fashion.

Yeah. I know when Bill Gross was on our podcast, he pointed out that this power-supply issue—pumped hydro is a huge way to store solar. Every piece of land on the planet that has a lake with a mountain next to it has already been purchased for this.

I was like, “What?” This stuff reacts so quickly. But the opportunity to make money on that actually came and went in 2024 and early 2025.

The other real estate—Alex was the first to point this out, actually—is the real estate that surrounds nuclear power plants, which normally nobody wants, right? You don't want to live right next to a nuke.

It's gone way up in value because that's the best place to put the data center, right? Because of power transmission costs and whatever other reason.

I think nuclear trends are coming really, really quickly.

I have a nuclear question. Why aren't we looking more at thorium, which is safe and a decent alternative? Worrying about uranium seems redundant to me.

Thorium—the work that Gates is backing in thorium reactors.

Yeah.

To first order, I think a good mental model is that everyone is looking at everything at this point.

Yeah. So, Alex, your investment advice—just to repeat it once again: this is not investment advice. We're just sharing what we think are hot areas that are going to create value in the world. Whether you invest or not is up to you.

Alex, so you're just saying, “Buy the QQQ”?

I'm not saying do anything or don't do anything. I am saying that my own outlook is that if you think the market is collectively superintelligent, it's difficult to rationalize trying to think that you have better insight into the market than the collective intelligence of the market, unless you have insider information.

This is a shrug-your-shoulders approach to investment advice, basically.

All right, let's move on. Dave, we ended this live on Leopold Aschenbrenner's holdings.

Yeah, let's just close this out real quick because I don't want to dwell on Leopold forever, but you can check this out at 13f.info. All hedge funds need to disclose their holdings every quarter, and so this comes right from that website. You can see a quarter ago he had $1 billion—you see it in the middle of the slide there. Now it's $2.1 billion. Some of that is inflows, and some of that is gains.

When you look at his holdings under the covers, you can see his CoreWeave investment went from about a $50 million investment to probably a $150-plus-million value. So he roughly tripled his money and made a $100 million-plus gain on CoreWeave alone.

Then, on the right, I asked Perplexity to try and simulate as best it could what he might own in Intel. His biggest holding, you remember, when we looked at this in August, was Intel, and it was listed as $500 million, which was half the fund, but it was options. That's the if-converted shares times the price. So what does he actually own?

It came back with this options ladder on the bottom right—these high-risk call options, which at the time were trading for under $1. I just went out and bought exactly what Perplexity said he must have. So now, today, I have the biggest one-day gain in my entire life. Actually, I'm partying today.

You're buying dinner next time.

Sure. But you can see in the bottom-right corner that, just today, those call options are up 255,250%. That was earlier this morning, so it's up a little bit more from there.

So brilliant. I mean, he saw it coming, he executed on it, and congratulations to him. We've got to get him on the pod.

Fantastic.

Well, more importantly, you called it out in July, Dave. So that's huge.

Yeah.

All right. You can't sit back and just watch today. It's about taking action. MIT grads help build neoclouds.

Yeah.

Back to you, Dave.

Yeah. I just wanted to throw this in there because I think it's cooler than cool. You have Kush Bhatia on the bottom right there, who's here at Link Ventures and has decided to co-found this company, working closely with Chase Lochmiller, who we had on stage last week in Palo Alto. Chase is in the chair on the top right.

Chase is building out Stargate. Remember, the $500 billion data center in Abilene, Texas? Peter, I think you committed for us to fly out there and record a podcast.

We'll do an episode of Moonshots from Stargate, sit down with Chase, and talk about, again, one of the most critical assets we have, which is power and data generation together.

Alex, are you going to come for that?

Absolutely. I wouldn't miss it for the world.

Yeah. Oh, that's so cool because I've seen some big data centers recently, but this thing is bigger than big. When you see things of that scale, it's kind of like a SpaceX launch: you just can't get it out of your mind ever again.

I noticed you didn't—you only asked Alex. You didn't invite Salim.

See, you're going to be there for sure. I have no doubt about that, although you're always bopping around India these days.

No, no, I'm back for a bit, but my schedule is a bit crazy.

So, what's a neocloud, buddy?

Yeah. A neocloud is a new-age data center built entirely for AI, from the ground up for AI. Some of them are local, so that you have low latency for things like voice AI, and some of them are central for training. That's what Abilene is: just a massive training data center.

These things are just going to get bigger and bigger and bigger as quickly as we can create the chips and the power supplies. The thing I really wanted to say about this story is that Kush's co-founder there is leaving his quant-trading job, where he's probably printing money, to do this because he absolutely hates quant trading.

You don't talk to anybody. You're not changing the world, and everything around you—all your classmates and everybody—is doing something world-changing. They're curing a disease, saving lives, building things that make the world smooth and seamless and entertaining, and you're just trading all day.

I love that. No offense to the quant traders out there—they're great—but I like the builders, and I'm passionate about the builders. I'm so glad that Kush was able to pull one of them away and put him into build mode.

So they're going to build, basically, a neocloud. It's a company that will use a combination of crypto and options—CBOE-listed options—to help neoclouds fund massive build-outs to create more and more of these things. That's all I can say about the plan so far because it's still in stealth.

It's fascinating. We're living in a world where we are tiling the world both in energy and in data and in intelligence. Ultimately, we're literally—I think about this—turning electrons into cash with Bitcoin and crypto. We're turning electrons into intelligence with our data centers, and it's accelerating. I can't even imagine where we're going to be a year from now, let alone 5 years from now.

Let's jump into the AI wars. Okay, here we go. DeepMind and OpenAI achieve gold at the coding Olympics. Over to you, Alex.

Gosh, this is such an exciting development. If you look at the past 6 or so months of developments from the frontier labs on the International Mathematical Olympiad, or IMO, golds—which several American and Chinese labs claimed credit for with their models—and the International Olympiad in Informatics, or IOI, it's absolutely incredible to see.

Once upon a time, I was on the U.S. Olympic team for the IOI, and now for the ICPC, which is a collegiate coding Olympics. If you think back to the original—

Q*. This was the reasoning model from OpenAI that was then renamed Project Strawberry, and renamed from Project Strawberry to the o-series of models, o1 and o3.

That was, in some sense, the last major overhang to collapse—the last major jump in capability from the frontier labs: the jump from models that couldn't reason to models that could reason.

This smells to me like the next major leap after reasoning models: models that are able to reason without having been trained on easy-to-verify problems. If you're solving a math problem on the IMO, you don't have the luxury of being able to take a partial result, a partial solution, and go up and ask a judge, “Am I correct?”

Similarly, here in these coding Olympics, we're starting to see AIs that are strictly superhuman, that do better than all of the humans in the competition on tasks that are difficult to verify in the middle of the contest.

My prediction here is that sometime, perhaps by the end of this calendar year, we're going to see the next major Strawberry, Q*, or o1 moment, when some of these models start to get broadly deployed and generally available. And, Peter, my model hypothesis regarding the future is: first, we solve superintelligence, and then we use superintelligence to solve math, science, and engineering.

I think this is “solve everything.” I think this is the key building block—the war cry. That's humanity's war cry, or AI's war cry.

So, Alex, a question for you. Dario Amodei, I don't know, 6 months ago went on record saying he expected we'd be at 90% of everything being coded by AI, and in 2026 we'd reach 100% of all coding being done by AI. Your thoughts on that?

Depending on which rumors you believe coming out of the frontier model developers—the frontier labs—it seems like we're very much on trajectory for that. You read stories of how now almost every frontier lab has its own agentic code generator. You read stories about people in these frontier labs yelling at their agentic code generators to do their work for them, rather than writing code at this point themselves. I think we more or less found our way into that part of Dario's future.

Incredible. Can I throw out a metaphor here, Alex, and tell me if you agree with this or not?

Sure.

When we started having AIs create art, they were very much poo-pooed at the beginning, and then over time, if they were anonymous, they started winning art competitions, right? Art is really generative at its core. When you're a high-functioning software developer, you're essentially an artist. You're pulling together libraries and trying to figure out how you're going to solve particular problems. It feels to me like that same tipping point has been achieved in coding, where they're using constructs that most human beings wouldn't consider, but they're bringing them to bear on a particular problem and solving it very, very elegantly. Would you agree with that metaphor?

It's an interesting question that I want to construe as a question about benchmarking. With the early generative art, arguably, other than sort of binary taste tests between competing models in various arenas, to my knowledge there wasn't really a benchmark for superhuman art generation. Whereas for coding, for math, and for some of these other relatively straightforward-to-verify problems and tasks, there very much is a goalpost—a benchmark, a waterline, if you will—that one can use to say, “Oh, this is superhuman versus subhuman.”

Which is why we've talked in the past about how we just zoomed right by the Turing test. Even though the Loebner Prize, rest in peace, isn't even active anymore, we just zoomed right by superhuman conversational agents. Why did we do that? Arguably because we didn't have a strict, rigorous benchmark in place to be the red-flag sign that we've just passed it.

The same goes for our earlier conversation. We're like, “Well, the benchmarks are all gone now, so what do we do?”

Yeah, right.

We need harder benchmarks, and we'll talk about that. I think benchmarks give us targets to shoot for, and they allow us to measure our progress. We humans do our best work when we're competing against each other. I'm going to say perhaps future agentic AIs are going to do good work when they're motivated to compete against each other, too. We'll see.

All right, other breaking news just in the last 48 hours: new Meta Ray-Ban Display glasses and a Neural Band. This is the first-ever private in-lens display controlled by a wristband, where you're using basically finger movements and picking up the EMG—electromyographic signals—coming out of your forearm to be able to type silently, right? So, you're talking to your wife or girlfriend or boyfriend, and you're having an in-depth conversation while responding to your text at the same time. What could possibly go wrong?

Let's take a look at this video. One moment. We can see the graphic here. You see the person's hand basically manipulating the display. You're looking off into the distance and typing at the same time. Thoughts? First of all, let's talk about the fashion statement here. What do you think of the glasses? They're a little bit too thick on the rim.

Good enough. People are going to wear them. They're close enough. The prior look-through goggles—no—was never going to happen. These are close enough.

Actually, if you think about how your phone got smaller and smaller and smaller, and then somebody said, “You know what? I actually want more screen space,” and now they're getting bigger, I think the glasses will do the same thing. As people get used to them, you're like, “Well, I want a little more battery life,” and they'll be fine.

At the Abundance Summit, I don't know, 5 years ago, I had a company called Mojo Vision there that had basically created a display on contact lenses, which I think would be super cool, and I think we'll probably get there eventually. But Zuck is predicting that these glasses will replace phones. The challenge I have is that I lose so many of my sunglasses all the time. I'm concerned about losing a $1,000 piece of equipment.

You know, what's really eye-opening for me is that if you want to grab a picture, you just hold your fingers like that, and that takes the picture instantaneously and then files it.

Yeah, I was like, “Oh my God, these gesture controls are going to get really ornate.” You'll see people at the airport waving their hands around, like, “What is going on there?” But it's like, “Take that picture, file that picture,” or whatever. It's a great way to control the machine. It's like, “Wow, I suddenly see where this is going.”

So, I love your glasses, buddy.

These are the Meta AI glasses you handed out a few years ago at Abundance, Peter. The first generation was surprisingly useful—really great to have conversations with and walk around, et cetera—as long as you're in the sun. If you're in the dark, it gets a little harder. I've got a prediction that this is a very temporary thing. I think we'll get so many more efficient ways of getting signals from noise out of the body, like that recent development where you could mouth what you're trying to say and it would just speak it for you, that I think this becomes irrelevant before it gets traction.

You're saying that the EMG on your wrist—the Neural Band—is going to become irrelevant?

Yeah, I think we'll find much more elegant ways of getting straight from the brain to what you're trying to do, rather than going through physical manipulation, et cetera.

Alex, what are you thinking about these glasses?

I tend to think that the glasses are lovely, but they're also a bit of a red herring. The input methods, I think, are far more interesting. It may start with this Meta Neural Band—a wristband that looks like a smartwatch and is effectively directly interfacing, albeit non-invasively, with the peripheral nervous system. But this, here in plain sight, is the beginnings of, I think, a brain-computer interface.

You start with the peripheral nervous system interface around the wrist. It's going to work its way up to some sort of extension of smart glasses, probably, or a headband that does a non-invasive brain-computer interface. Vernor Vinge wrote about this extensively in Rainbows End and other novels. We're finally living in the future where silent messaging that results from non-invasive peripheral nervous system computer interfaces is a reality. This probably ends up being the first mass-market, effectively BCI.

Yep. You know what I find exciting?

There's a visual in my head of it slowly crawling up closer and closer to your brain as it gets smarter. It's a great horror movie waiting to happen.

Injecting nanites through the ear to connect to your temporal lobe. What I find fascinating here is a future in which we are constantly uploading our visual input, right? Imagine that we have enough memory and enough bandwidth—which we do—and we're dribbling bits all the time. Whatever we've seen, whatever we've heard, every interaction we've had is available to your AI and available for a full recall on the cloud.

How many times have you had a conversation with your spouse in which you said, “No, no, no. I didn't say that. I said this”? Well, you're never going to have that argument again. It's all going to be on record.

But Peter, also, I think even putting aside the various Black Mirror-esque scenarios enabled by that, this is going to be amazing—amazing—for fleet learning and training of humanoid robots.

Yeah. Imagine a world where a billion-plus of these are deployed, and now you can do fleet learning at scale for every sort of long-tail occupational manual-labor task, just from first-person video data coming out of these smart glasses. And that's what Elon said about training up Optimus 3. We're no longer going to do it in full-body suits where we're tracking the movements and 3D position of the fingers and hands. We're going to be doing it from just visual learning, in the same way that self-driving was picked up from visual learning.

This is also going to be of massive value to the elder population, people who have degrading memory. I simply want it so that when someone's approaching me, my AI does facial recognition, calls up the last conversation, and reminds me to ask, “How's your son or daughter's birthday?” I mean, it's going to create a very rich level of social interaction.

Independent of the gesture stuff, the broader implications of this are so huge, right? It's like all the stuff that you talked about, Peter: the cognitive enhancement, the memory enhancement. I think it's just the interface that needs to be solved.

Just a humorous memory here. When you were talking about marriage, Peter, and trying to recollect conversations, I remembered coming back from one of my business trips. I was a bit disheveled and jet-lagged, and she literally looked at me and said, “The user interface around you right now is really bad.”

I’ve never forgotten that.

Oh, I love that. For those listening, this is a window into what’s coming rapidly. We’ve been saying for some time that the form factor of the phone is going to disappear. It’s going to come into headwear. It’s going to become—

Yeah. Go on.

It just occurred to me. Based on what Alex just said, I always wonder: Why do I look forward to this podcast so much and cancel all my meetings to do it? What Alex just said made me realize something I hadn’t realized before.

When the laptop first came out, everybody wanted to attach a mouse to it because they were used to using a mouse. Then the touchscreen on all the iPhones came out, and everyone got used to swiping, so they backported the trackpad to the laptop. Now everything works the way it does on your phone. You swipe left, swipe up, down, or whatever.

The new interface is going to be waving your hands in the air because that’s the only way to control the glasses. Then people are going to fall in love with that, and the keyboard and trackpad will go away because they’ll say, “Why can’t I use my gestures on my laptop? It’s got a camera.” It totally didn’t dawn on me until just now, based on what Alex said, that’s how it’s going to evolve.

Yeah. This is going to be a driver for increases in BCI. At the Abundance Summit this year, I’ve got 2 of the top brain-computer interface companies coming. One is invasive and one is noninvasive.

I think our ability to get the signal out of the noise is going to enable us to think and say, “Click picture,” or “Pull up my schedule for the day.” I think this form of user interface is ultimately going to drive a need for rapid BCI capabilities, because I don’t think you’re going to want to use your hands for the long run. I think you’re going to want to think and Google.

It’s only a phase. To the extent that you’re a technological determinist and you think there’s a sort of preordained sequence of technologies that’s likely to happen, I think the consensus among the futurist community is that we go from smartphones to smart glasses to BCIs to some sort of human-machine symbiosis or uploads, in roughly that order. If you believe in that sequence, then this is right on the money for where we’re supposed to be.

Yeah. And let’s note that the other eye-opener is that this is a long progression that’s been going on for hundreds of years. Dave, you sometimes wear glasses. We would call you a transhumanist because you’re augmenting yourself with technology.

The minute you get a vaccination, you’re a cyborg, technically. So this has been a long thing. We’re just accelerating the pace of it very rapidly.

And we’re allowing ourselves to utilize more and more extreme variations of this.

Yeah. The other insight that comes out of this brainstorming we do together is that a lot of people I talk to say, “Look, I don’t need my computer right in front of my eyes all day, and I need to disconnect from it.”

That’s true, but as soon as it’s superintelligent and giving you crazy-great advice about where to stop and eat, where to drive, what to do next, and how to live your life, then you get addicted to it so quickly. Right now, anyone under the age of 20 has iPhone separation anxiety, or phone separation anxiety. If you try to take away their phone for a weekend and they go camping in the woods without it, they’re just in shock.

But once the thing is superintelligent, being disconnected from it will basically make you naked. Your symbiote is gone.

We made a decision in our family not to get our kids, who are 14 now, phones until at least age 16. It’s so strange because when I go to a birthday party for their classmates, at age 14, they’re all heads-down in their phones, ignoring everything else that’s going on. I just hate that.

It’s been tough. But it’s harder than not giving your kids phones; it’s taking them away after you’ve given them to them. So, I’m hoping the kids will skip the phone and go directly to a BCI. We’ll see.

I think this is so wise, Peter, because at this age, their brains are just finalizing the construction of the neocortex, and to have that diverted away is very, very damaging.

I mean, just to be clear, they’re not saints, right? They’re on Roblox and Minecraft and all the games on there. They do have MacBooks, which they like, but you’re not carrying a MacBook around during social interactions, during a baseball game, or during a birthday party. So there’s some level of balance.

To other parents out there, if your kids don’t have a phone yet, consider not giving one to them. The data is so strong about mental health.

I think that’s the key point. There’s so much strong evidence, and I think one of the things we’re really deep on in this podcast is evidence-based framing of discussions.

Yeah. All right, let’s move on to our next topic here. This was a tweet that went out this morning that I flagged: “The U.S. risks losing to China in chip wars.” This comes from our U.S. AI and crypto czar, David Sacks. Let me just read it:

“Huawei just launched a new AI chip, and China has ordered firms to stop buying certain NVIDIA chips. China is not desperate for our chips. It’s producing its own and intends to compete globally in the semiconductor marketplace. If U.S. firms are blocked by excessive controls, we risk forfeiting the AI race to China. The hawk position—to help American companies win, not help Huawei build a Digital Silk Road—is so critically important.”

We’ve talked about this. As soon as you make anything illegal or stop trade, you’re incentivizing the other side—in this case, China—to develop the capabilities and then compete directly with you. This is about who’s going to buy throughout Europe, throughout Africa, and throughout the rest of Southeast Asia. What chips are these companies going to buy? Is it Intel and NVIDIA, or is it Huawei? Alex, your thoughts here?

I think it’s instructive to look at what happened to computing during the Cold War. In particular, there was so much siloization in the Soviet Union of its computing stack. They were inventing all sorts of exotic architectures, like ternary computers, while in the West—and ultimately in the world—we settled on a binary architecture, zeros and ones, for computers.

The Soviet Union, in part because of the Iron Curtain, was experimenting deep in its tech stack with all sorts of exotic architectural options that the West wasn’t considering. If the situation does play out in a way where there is effectively competition deep into the tech stack between great powers, then based on history, I would expect to see, ironically, greater diversity deeper in the stack.

Maybe not quite at the level of binary versus ternary computing systems, but certainly I would expect to see more experimentation deep in the stack. Does this ultimately help America by having a great, capable competitor out there?

Well, I think there are different sides one can take in the broader discussion. I know, Peter, you’re a fan of the alternative-reality television series For All Mankind, which explores what would have perhaps happened if the Soviet Union had been the first to land a person on the moon rather than the United States, and all of the follow-on consequences as that original event ripples throughout the decades.

It’s difficult to have a crystal ball to know what happens decades from now, depending on how the great-power competition over the semiconductor tech stack plays out. It’s difficult to know for sure without a crystal ball.

Yeah, it’s hard to predict the future. I get that. But what I’m saying here is that if the U.S. was so dominant that it was the only player, the incentive to continue innovating would still be there, but it would slow down somewhat.

When you’ve got Huawei leveling up to the level of NVIDIA, I think you’re going to see the White House and capital in the United States really doubling down. So, is this reality about Huawei and China going to become additional fuel for the U.S. government and U.S. industries to align? Again, we’re in a war footing here. This is a war cry that we just heard from David Sacks.

Yeah. I think what’s new and interesting to me about what David’s saying here is the Digital Silk Road. We’re not going to lose to China if it’s like the U.S. versus the Soviet Union. But if the whole world ends up partnering with China and it’s the U.S. versus the world—and when he got back from India, you said, “Hey, what’s the attitude like in India?” and Salim says, “It’s like America”—I’m like, “What? Where’s that coming from?”

If Huawei becomes the open and friendly partner to the world and the U.S. gets isolationist, that’s what I think David is warning us against here. What is Europe going to do? What is India going to do with its 1.4 billion people?

I think this is more about the politics than an innovation race, because there’s so much incentive in the world to turn away from the U.S. right now, given the tariffs and other things. That’s the forcing function, rather than the chip wars themselves.

It's a pretty bad situation right now. It's 40 years of U.S. diplomacy trying to keep India and China apart, and now that's gone. I think there's a bigger game at play here. There's also—and I would love your comment on this, Alex—the whole open-source movement, right? When you look at different countries, everybody wants their sovereign AI, and everyone wants data centers.

Who are they going to use? Whose AI engines? Whose chipsets are they going to use? I mean, the world is going to split between Huawei and Chinese open-source models, and NVIDIA and U.S. models. We've seen in the last few WTF episodes—we've talked about OpenAI going aggressively into India, the U.K., Greece, and other places, right? Because if you embed yourself there and people grow up with your models, they're likely to stick with them. I think this is a land grab in a lot of different ways at a very critical time over the next 2 or 3 years. Alex, are you going to tell me I'm wrong, or do you agree?

I'll half-agree. The world is hungry for intelligence in general, and maybe superintelligence in particular. But I also think there's a sense in which accelerated computing is the geopolitically significant scarce resource of the moment. If we were having this discussion 10 years from now, maybe we'll be bellyaching over, "Oh, you know, the Mars colony is running out of resources," or, "The lunar colony has a lot of competition on the Moon."

So I think, in some sense, it's important to jump up a few levels and abstract out the principle that, as long as we're in a pre-general-abundance world in civilization, there will always tend to be, by power-law statistics alone, 1 maximally geopolitically important and scarce resource at any given time. At the moment, it's accelerated computing.

Here's our next piece of news. It's fascinating. It's part of our "Solve Everything" segment. Alex, I'm going to ask you: AI discovers a new solution to a century-old fluid dynamics problem. How critical is this? How valuable is this? Why should people care?

Oh, it's incredible. We've spoken in the past, Peter, about this notion of AI solving everything in math, science, and engineering. There are million-dollar prizes for solving particular problems in math, science, and engineering, and this development, which was launched literally a few hours before we recorded this episode by Google DeepMind, is one such task. So this is progress toward a general solution to a million-dollar problem.

It's one of the so-called Clay Millennium Prize Problems. It's referred to in math and mathematical physics as the Navier–Stokes problem. And the problem is basically: if you have a fluid, does the fluid ever have infinite densities or other unnatural properties? Is it possible to stir a fluid in just the right way that a singularity appears in it?

The approach that Google DeepMind has taken is really, really interesting. For those who haven't followed the Navier–Stokes problem—which I would say, parenthetically, is a key signpost that we're on the verge of solving math when AI is taking these grand challenges in math and just knocking them out—it's a constructive approach. Mathematicians like Terry Tao and others have been knocking on the door of this problem for a decade-plus, and the proposed solution to this problem, which appears to be a construction of a way to create singularities out of fluids, could, as Terry and others have pointed out, enable us to build nanomachines and computers out of just pure fluids.

Imagine if you could stir a glass of water in just the right way such that it formed circuits, and those circuits performed general computations. And even more exotically, imagine you could stir a glass of water in just the right way such that it created nanomachines out of fluid-flow patterns of turbulence in that glass of water that created copies of themselves.

So imagine creating self-replicating fluid nanomachines out of a given fluid if you could just position the initial velocities and densities of the fluid in just the right way. If there's a constructive solution for Navier–Stokes that results in singularities, as Terry and others have studied this, it could literally unlock our ability—and granted, this is on the more exotic end of applications—to build self-replicating fluid nanomachines.

Wow. I know that sounds far-fetched.

If you go to alexwg.org, there are a whole bunch of essays that Alex has written on computing substrates. People used to think about this a lot, and, as you mentioned, Alex, back in the Soviet era, when computing was new, people thought about this all the time. Now everyone takes for granted that it's all going to run on NVIDIA chips, but the neural algorithms—the new algorithms that matter—are just a couple hundred lines of code. All the complexity is in the parameter file and the self-organization, so the idea of computing substrates that are self-organized is very viable with these new algorithms. It's not as far-fetched as it sounds at all.

And to Dave's point, maybe just to build on that, at the moment we build our GPUs based on silicon CMOS. Maybe at some point in the future, if Navier–Stokes and other related problems have constructive solutions, maybe we'll be building our future accelerated compute out of plasmas or out of fluids. Even a cup of coffee might be more powerful computationally than an off-the-shelf NVIDIA RTX.

It reminds me of The Hitchhiker's Guide to the Galaxy, where the whole world is essentially 1 big organic computer.

The key is asking the right questions. Yes, I love it. All right, next item here on our docket today: "Study shows AI model can forecast 1,000-plus diseases." Researchers built Delphi-2M, which can predict how a person's health changes over time based on their medical history. AI models can forecast risks for over 1,000 different diseases and simulate what someone's health might look like in 10 to 20 years, shaping how doctors and health care systems plan to treat and prevent illness.

We're reaching this singularity in health. We've seen Demis predict the end of all disease within the next 10 years. We've seen Dario Amodei talk about doubling the human lifespan. Again, all of this is happening not because we humans are becoming smarter or that our biology is getting simpler. It's just that we're able to use advancing AI and soon ASI to understand what's going on in your 40 trillion cells, running 2 to 5 billion chemical reactions per second per cell. Who wants to jump in on this? Alex, you want to jump in again?

Yeah, I'll take this one. This was a lovely paper in this week's Nature from a German team that I think underlines that these foundation models that we're training aren't just about text. It's not just text models or image generation with Nano Banana, as we've discussed previously, or even video or audio. Now this paper is widely popularizing the notion of literally disease as its own modality for foundation models that can be trained with disease-token sequences.

Just like a sentence is tokenized into a sequence of almost word-length tokens and then fed to a frontier model like ChatGPT, this is a model that tokenizes the electronic medical-record history of many, many patients into disease tokens and then can reason over a person's entire history of health and illness as if it were a sentence in some language.

The upshot of all of this is that, as with a conventional frontier model that does next-token prediction in text, it can predict the next disease token that a person is likely to incur. At some point, you get a sort of medical time machine that's able to extrapolate out entire contingency trees of a patient's future medical history. And as they say, "If you tell me where I'm going to die, I'll make sure I'll never go there." If you can extrapolate out a contingency tree of a person's entire future history of health, you can make sure to avoid branches that you'd rather not spend time on.

That's incredible. The sequence of words you put forward here—if you went back in history 5 years and spoke them, like "disease-token sequence," people would say, "What in the world are you smoking?"

What struck me here was the tokenization of not just text, but the medical history—what Alex is referencing. The second thing that struck me was that this was done on a population of close to 1.9 million people. This is a big number, and it suggests that diseases can be modeled across large populations in a really powerful way.

This is incredible.

Yeah.

That's incredible, and I encourage people not to be intimidated by it either. We're seeing more and more business plans from people who are building topic-specific foundation models, and the funding is there. The people who jump in are succeeding wildly because it sounds really intimidating—because it sounds ultra-complex—but AI is such an incredible assistant, and the underlying algorithms are very understandable if you're willing to jump in and study.

This seemed to me like one of those kinds of things. If you look at the previous conversation about computational structures in a glass, if we solve that problem, it's a little bit down the line. This seemed to me immediately usable because any single person could take their medical history, throw it into this thing, and say, “Okay, project out where I'm going to be.” Am I wrong on that?

No, you're not.

You can go to your Gemini 2.5 voice AI or OpenAI's GPT-5 voice AI and have a conversation and say, “On the WTF episode, there was this notion of Delphi-2M about forecasting diseases,” and then have a conversation with the AI and go deep. It has this data in it, so ask, “What does that mean, and what are the implications for me?” I think your ability to have follow-on conversations after this podcast, to dive deep on a particular element, is powerful. Alex is our own personal AI on this podcast, but you can imagine—

Yeah, for sure.

You could also use it to say, “Hey, watch that episode and tell me where I should invest.”

Yeah, not investment advice. I'm going to say that again. I think this is really exciting. So, at Fountain Life, when I do my upload, I get 200 gigabytes of data, and I have that data on my phone uploaded into our Zori AI, which allows me to interpolate on my current medical information. I think what this is saying is that it's going to enable me to extrapolate beyond my current medical information.

Yeah, for sure.

If you take that data from Fountain Life, Peter, and map it onto this, right there, you can get instant usage out of it, instant value, right? Yeah, for sure.

At some point, with the ability to extrapolate contingency trees for health, my expectation is that healthcare becomes almost like a game of chess, where you're playing against an opponent—biology, or at least uninterrupted biology—and the goal is to win. If you can map out, to high confidence, the contingency tree of interventions, those would be like game actions, and the underlying course of biology without that interaction—the goal of winning—is longevity.

Yep, absolutely. I'd like to move to our next category, which is robotics. A lot of news in the robotics world this week as well. Let's jump in.

The first piece is a friend of the pod, Brett Adcock, who's been with us a couple of times. He was on my stage at the Abundance Summit last year. This year, our theme at Abundance Summit 2026 is “The Rise of Digital Superintelligence and Humanoid Robots.” I think, if successful, I'll have 5 different humanoid robot companies there, with their robots, where you can play and touch and feel them and see what it's like.

Brett just announced this week that he raised over $1 billion in a Series C at an extraordinary valuation. Good on Brett—$93 billion. Full disclosure: I am an investor in Figure through my Bold Capital Partners venture fund. The investors include NVIDIA, Intel Capital, Brookfield, and Parkway Venture Capital. Thus far, Figure has raised $2 billion over the last 2 years, making it one of the fastest-growing startups. What's interesting is that Brett is investing this money really into manufacturing, right?

Another company we did a podcast on—Dave and I were up in Palo Alto with Bernt Ørnich, the CEO of 1X Technologies, which makes NEO Gamma. They're in the midst of their raise right now, and I wish them the very best of luck—an incredible company. They're going to be investing their capital, once again, in building out manufacturing. We're not at the point yet where it's going to be robots building robots, but we'll be there in the near future.

Dave, do you remember the conversation we had with Bernt about sort of the iPhone-scaling goal that he had?

Of course. Of course I do. He had it all mapped out brilliantly in his mind, actually, and he was a phenomenal guy. Brett Adcock is, too. Brett Adcock is the most down-to-earth, salt-of-the-earth, Midwestern, snow-shoveling guy made good. So, I'm really, really cheering for both of these guys.

They're both wrestling with the supply chain of parts. The supply chain for robotics parts is terrible. China's way ahead of us on that front, by the way, but that has got to get figured out really fast. It just takes capital and focus, and now they have capital, so I think it'll get built out very, very quickly. Humanoids first, and then, as Salim has been saying, it'll branch out from humanoids very, very quickly.

I thought it was kind of funny. Bernt made a big deal about the soft skin and the friendly look. Do you remember the shirts they gave us? They make their own fabric—super comfortable.

Yeah, I love it.

Yeah, your robot is better dressed than you are. Brett is going down the Terminator look here, which—I don't know what the difference in philosophy is driven by—but if you look underneath, the 1X robots that we saw in Palo Alto look just like this. They're very Terminator-looking underneath.

Exactly. It really is. 1X's robots are built to be soft and approachable—almost huggable, I would say. But what I found so impressive about Brett Adcock is that he's a consummate engineer-founder. He was running Archer Aviation, one of the flying-car companies that went public for multiple billions, and when Brett left Archer to start this, he committed a couple hundred million dollars of his own capital to get it going.

That ballsy move enabled him to hire extraordinary talent from around the world and really uplevel, level after level. I was lucky we got Bold Capital in very early.

It's so cool to see people like this succeed, too, because I had a great conversation yesterday with 2 MIT sophomores in Course 6-5, which is chips and software meshed together, and they're so excited about the physical world.

For the longest time, there was no way for people who liked physical things to become this successful. You had to do software. You had to do software. And that was fine. Elon was the only one.

He was the only one. Now, actually, Elon has inspired a whole generation. These sophomores are wicked excited to build physical things using all this AI technology.

I think there's something else here, though, that's worth pointing out: these folks—Brett and others—are not interested in the money. They're interested in making a massive difference and transforming the world.

Yes. Exactly.

When you have that purpose, and then you can have a success and deploy that capital back into the positive feedback loop and flywheel that creates, it's inspiring. Elon has done that, inspiring the next generation, and Brett will inspire several more generations. It's incredible to see, and it's hard to become pessimistic about the world when you see this.

Yeah.

Yeah. I totally agree. I mean, there's this feeling—you see it in the posts, actually, on the podcast—of just fear and fear and fear and fear. A lot of the people that rose to the top were kind of whack jobs. There's no doubt about it. But then you meet these guys that are the new generation—

They're so focused on the good of the world, and they're so inspiring. So, it does make you feel confident that—

Good things are coming.

I have a funny story here. I remember meeting one of the founders of one of these Layer 2 blockchains, and I said to him, “Why would you do this on Ethereum when you could build your own Layer 1 blockchain?”

And he goes, “Well, according to your own book, a new innovation has to be 10 times better than the market; otherwise, the market ignores them.”

So I'm like, “Holy crap, this little 18-year-old guy is quoting my own book back at me.” It was so impressive. This whole new generation is all like that. They're so with it. It's incredible. It's very annoying that I wasn't that alive now at 18 to take on this kind of crazy risk.

Don't worry, buddy. We're going to give you an extra 50-plus years of lifespan here.

Oh, yeah. I have to get myself to Fountain Life.

One thing that's important here is that people are hungry for optimism. People are hungry for a positive, optimistic, hopeful future because they're so decimated. I had 2 tweets go viral and get like 5,000 to 8,000 likes within 2 days, and they were both about optimism. It helped that Elon retweeted both of them, but people want optimism. They really do, because we've had so much negative news on the planet.

Elon, you said in your tweet that the only time more exciting to be alive than today is tomorrow, which I thought was great. Elon said, “Better to live life,” as a retweet of yours saying, “Better be optimistic and wrong than pessimistic and right.”

I would go further and say, even beyond optimism. It’s difficult for me, at least, to imagine a future where real growth, real economic growth, and the wealth of our planet grow significantly without passing through a humanoid robot or some other substantially similar robotic stage where manual labor is scaled up by orders of magnitude. It’s almost an instrumentally convergent goal, I think, for achieving some of the more ambitious futures.

I want to remind people: your mindset is your neural net, right? The way—what you see, what you hear, who you speak to, the way you see the world—is constantly shaping the way you react to opportunities and challenges. You need to surround yourself with people who see the world in a positive, can-do, optimistic, abundance-minded future, which is why we do this podcast, why we spend the time, why I love my moonshot mates so much. Hopefully, this is helping you shape your purpose-driven, abundance, exponential mindset as well.

All right, you can’t talk about robots without talking about what Elon is doing. He has teased Optimus 3 and provided some predictions. So, in this image, if you’re watching this on YouTube, which I hope you are, there’s an image of Optimus Gen 2 and a vision of Generation 3, much sleeker. It looks like Optimus went on a little bit of a diet and was trimmed—

Sticking GLP-1s.

Here are some of the details: the first-ever OLED face display for expression and stronger hands. I think one of the things we talked about, Dave, with Brett Adcock, was the whole element of the face and how close to the uncanny valley we allow it to go. Walking speeds are closer to humans, with better balance and navigation. Prototypes by the end of 2025, mass production in 2026, with a goal of 1 million robots per year. I’m not sure if that’s for 2026 or 2027, but it’s a $25 trillion market.

As Elon has said publicly, he expects this to be the reason that he gets his pay package. Remember, he’s got a $1 trillion pay package if he can grow Tesla to $8 trillion, which is an order of 8–12× where it is today. Thoughts on this, gentlemen?

I think colonizing and developing the solar system is going to entail lots of dirty, dull, and dangerous jobs. So, for one, I’ll be delighted to live in a near future where the solar system is populated by lots of humanoid robotics of general AI capacity.

On the other hand, going back to the discussion we had last time, at some point I would predict the personhood discussion is inevitably going to happen. We’re just trotting into various sci-fi futures. At some point, if we reach some threshold of AGI personhood, and AGI is embodied in human form factors or humanlike form factors, someone, somewhere will surely want to push on the personhood question. If the mature embodiment of these vision-language-action, or VLA, models ends up becoming more and more humanlike, I think that will push the discussion even more.

I have a question for the 3 of you. The best conversation around personhood in this that I’ve seen was Star Trek: The Next Generation with Data. We had several episodes talking about rights, humanity, and emotions. He had an emotional chip embedded in him. Has any of you seen a better treatment of the personhood question?

Bicentennial Man did a good job.

Oh, in sci-fi.

Yeah.

Data’s got to be the best, though, right?

Yeah.

That’s what I thought. I would say Accelerando. Yeah, so, Alex, by the way,

I know everyone should read that. You also mentioned Vernor Vinge’s Rainbow’s End and anything else that comes to mind. You know, the vision and terminology from those books—please read the book, or at least skim through it, to get on the same page with Alex. It’s absolutely worth making the investment.

We should put a list of those in the show notes. Just to give credit here, Vernor Vinge was the first person to coin the term “the technological singularity.”

Actually, he popularized it.

The notion goes all the way back to I. J. Good and the intelligence explosion as a mathematical singularity, but Vinge was the one who first popularized the technological singularity as such.

Yeah.

My prediction was boots on Mars by 2030, except they’re going to be robot boots, right? I mean—

I hope they have wheels. I hope they have wheels.

No, they’re not going to have wheels, for God’s sake. They’re climbing on rocks and across Martian terrain.

Have them have both. Have them have both. You could have the optionality, for God’s sake.

Nope, nope. They’re going to help us do what humans do best: explore. But here’s what’s interesting, right? You send the robot fleets there first and they build the habitats. They get everything ready. We squishy carbon life forms don’t want to end up in a harsh environment. Let the robots go set it up for us, and then we’ll go join them. But this is—

Or hybrid solutions. Telerobotics. The problem is, of course, latency, but there are various compensations for that.

You know what I totally don’t get, and you can explain it to me if you can—and Peter touched on this a second ago. I had this feeling in my chest the other day of just this raw enthusiasm that I don’t think I’ve had since I was playing high school soccer and you’re about to win a game. It’s this physical enthusiasm vibe, and I was working on a neural net. It’s a neural net that’s designed to build other neural nets, and it just freaking worked. It just flat-out cracked the code and worked.

This is all part of this new quantum company that I’m working on with Alex. So you take that same thought process and you say, within robotics, all 3 of these companies are working on humanoid robots, but wouldn’t the first thing they want to build be what Peter said a second ago? The robot that’s manufacturing the robot is far more important than the robot itself because of the self-improvement effect.

I don’t know if they’re doing it quietly and not showing anybody. When you talk about 1X, Figure, and Tesla, I haven’t seen them talk about that at all. But that’s the first thing I would do: build the robot that’s manufacturing the robot.

For what it’s worth, I think Elon has made several comments about the importance of “the machine that makes the machine” and has said on several occasions that some of the earliest deployments of Optimus will be in Tesla factories. To the extent that Tesla factories are turning out not just cars but also robots, I think that sort of self-licking ice cream cone, or recursive self-improvement cycle, is either already started or about to start.

Yeah. Well, for everybody listening, the important thing here is to realize that recursive self-improvement is now happening in AI and will accelerate, and will soon be happening in mechanical systems and will accelerate. This is where we have this intelligence explosion. I still feel like Ray has it wrong on the singularity. I think he’s too far out there. I don’t know—it’s like I feel a singularity every moment of the day. It depends how you define it.

I’d like to throw in the following comment: when you think about recursive self-improvement, that’s actually the fundamental basis of evolution. We’re just continuing a trajectory that’s been happening for billions of years—

Just a billion times faster.

Yeah, and accelerating.

To me, the singularity is like a nuclear reaction. It’s pretty much happening or not happening. It’s not like strong AI needs a definition and AGI needs a definition, but the singularity—it’s obvious when that’s happening.

Because the rate of self-improvement is just crazy. It spirals; it’s feeding itself.

It’s amazing.

I don’t think it’s going to be that obvious.

Now, remember: exponential growth looks vertical in the future, and it looks horizontal behind you. We’re kind of there now.

I think we’re going to see the singularity very soon, and you’re going to see an immediate 1,000× step up in the performance of neural network software. The results of that are going to be unmistakable: instant solving of diseases, instant solving of impossible math problems. It’s all going to happen within, say, a 30-day window, from self-improvement to the other side of that, with no change in the chips—just purely software advances.

But then you have a distribution problem, right? This is where William Gibson’s quote comes in: “The future is already here; it’s just unevenly distributed.” We talk often about Mad Max and Star Trek as 2 framings for the future. In an ideal world, you end up with one rather than the other. But we can see them both happening. Ukraine and Gaza are Mad Max, and we have Star Trek in many other parts of the world.

And so I think we'll end up with that uneven distribution, which is going to cause a lot of stress and chaos.

I don't want to end on a negative note like that. No, no. I'm just fundamentally optimistic because the positive will drag along the negative, right? We talk a lot about the fact that if you can lift the bottom billion—and now most of them have smartphones with AI built into them—I mean, the potential for humanity is incredible.

I think the important note here is that, in a world in which every mom knows her children have access to all the food, water, energy, healthcare, and education, you have more to lose by warring than you have to gain. If you're—no, seriously—if you're a population that has nothing to lose, I think that's where we get this negative.

But just to echo Dave's comments, I always feel so much smarter after this hour and a quarter that we spend conversing about this than before. So it's great.

Yeah.