达沃斯2026：中美AI竞赛、GPU外交与走上街头的机器人 | #225

A lot is happening at Davos—a lot of conversations that are important for us to recount here.

In past years, it was dominated by politicians and economic policy. Then it moved to the internet a little bit. This year, it’s all AI.

We are knocking on the door of these incredible capabilities, right? The ability to build basically machines out of sand.

Maybe it would be good to have a slightly slower pace than we’re currently predicting—even my timelines—so that we can get this right. Should we slow down? What’s the path humanity should take?

Where do you see the US and China right now in the AI race?

I still think that the US is in the lead. I think that our models are better and our chips are better, but they do have other advantages. They are spinning up power generation faster than we are. That’s one area.

If you believe that energy is at the heart of it and is the core of the inner loop, China’s going to go way ahead anyway. But I think the real differentiator in the race is going to be application-layer dominance, not frontier benchmarks.

The problem is that you always need a bad guy in a movie.

Now that’s a moonshot, ladies and gentlemen.

Dave and Alex, you’re just back from Davos, and that’s the theme of our show today. A lot is happening at Davos—a lot of conversations that are important for us to recount here. So tell us, what was it like?

It’s a lot of time zones away.

Yeah, a lot.

Actually, Larry Fink, who is the organizer this year, said, “We should do this in Detroit next year.”

How so?

That did not go over well across Europe. “We want to include more views from around the world. This exclusive resort town is just too swank. Let’s go to Detroit.”

Well, here are some fun photos of you guys. It was cold and sunny.

Is that Sandy Pentland?

Yeah, Sandy Pentland with his Brass Rat. Always great to see him. He’s been there since the dawn of Davos.

I would say there were robots on the streets, billionaires eating out of food trucks, and anti-aircraft guns on the ice pond.

Nice. And with Trump—

There were actually 3,000 more security personnel than in the past 6 years—3,000 armed people in fatigues with machine guns, extending halfway to Zurich, actually. I was surprised, but Donald Trump was there this year, so I guess they ratcheted it up. He attracts a crowd.

So, Dave and Alex, what was the vibe like? What did it feel like there? I mean, you’ve been there numerous times. I was there once as a speaker, and it was overwhelming beyond belief in terms of trying to find people there. It’s a zoo, isn’t it?

Yeah, it’s a zoo. If I were to characterize what was different this year, the quintessential example is that Alex was there. In past years, it was dominated by politicians and economic policy. Then it moved to the internet a little bit. This year, it was all AI—completely dominated by AI—which is encouraging.

No one had anything intelligent to say other than the usual suspects, Dario and Demis, and the people we see all the time. But at least they were listening: global leaders and presidents of pretty much every country, listening to a pretty much all-AI dialogue. That gives you some hope that people are beginning to get ready. The fact that Alex was there is kind of a bellwether of where it’s likely to go.

So, Alex, do you want to take it away?

Yeah, it was, of course, amazing, and there really were robots in the streets. A couple of frontier labs had houses there. Imagine, for those who haven’t had this experience, almost a World’s Fair or World Expo set in the Alps, with major governments having their own houses—literally taking over storefronts, restaurants, and convenience stores.

If anyone wants to make a killing at Davos next year, set up a restaurant. It is nearly impossible to find good food short of food trucks at Davos. Someone will make an absolute killing setting up a restaurant that’s still a restaurant during Davos week.

Imagine a World’s Fair with the governments, the frontier labs, and some of the major corporations and tech companies all on an equal footing, all hijacking—or, in Invasion of the Body Snatchers style, consuming—the storefronts of an Alpine resort town. That’s Davos.

I had an amazing experience. I moderated something like 8 or 10 different events with OpenAI executives and DeepMind executives. I did a fun panel with Llion Jones, one of the co-creators of the Transformer architecture.

You can see here one of my photos with US Under Secretary of State Sarah Rogers, along with some absolutely amazing and, hopefully, fruitful discussions about the era of GPU diplomacy that we find ourselves in. I had discussions with Jack Hidary of Sandbox AQ and Daniela Rus, the head of MIT’s Computer Science and AI Lab.

I think it was an absolutely incredible experience, but I also think, to Dave’s point, that AI is the story. It’s no longer a world where governments get together to talk about governing the governed. It’s now, I think, a world where AI and superintelligence are the story of the world economy.

Walking down the street and seeing the DHL robo-dog walking around, or humanoid robots taking a stroll from AI House, exemplified what we’re seeing in the global economy on a broader scale.

And that’s a good transition to looking at a little bit of the content from Davos. What we’ve done is cherry-picked a number of conversations. There’s a huge amount of this flurrying through X and various platforms. We’ve chosen a few to bring up here, listen to, and discuss amongst our moonshot mates.

Let’s begin with Dario Amodei, the CEO of Anthropic. He was a rock star during Davos, as was Jensen Huang. Let’s listen to Dario first on the economy.

If you look at what AI is capable of, and you have these models that are getting more and more capable across a wide range of cognitive tasks, you look at all labor in the economy—that’s something like $50 trillion a year. I could easily imagine that the revenue of the industry, or even single companies, if it’s even 10% of that, could be $5 trillion a year.

That’s something we haven’t seen in the history of the world. That creates all kinds of problems, as well as creating all kinds of growth.

The largest infrastructure buildout in human history. We’re now a few hundred billion dollars into it.

That’s it.

We’re a few hundred billion dollars into it. Larry and I get the opportunity to work on many projects together. There are trillions of dollars of infrastructure that need to be built out, and it’s sensible. It’s sensible because all of these contexts have to be processed so that the AI—so that the models—can generate the intelligence necessary to power the applications that ultimately sit on top.

We have chip factories, computer factories, and AI factories all being built around the world.

I saw a text this morning from Elon saying he expects to see $100 trillion company valuations coming up in the next few years. I think he said 2030. When asked which company it was—

The combination of SpaceX and Tesla.

So, a trillion here, a trillion there. What are you thinking about these numbers being thrown around in the economy?

I start to look at this as becoming meaningless. In an abundance environment, whether it’s $10 trillion or $100 trillion, the whole thing becomes arbitrary and is not a real gauge of strength.

What I found really interesting about the NVIDIA and Dario conversation was that they’re on opposite ends of the stack, and they’re really saying roughly the same thing, because compute is becoming the new oil and the new electricity. This is really just incredible stuff. It was amazing watching the output of the overall event from a distance.

Well, just in terms of tone, Jensen sounds like he’s talking to kindergarten. He’s talking so slowly, and the things that we talk about on the podcast—you are 100 times more detailed and sophisticated. That gives you a sense of the audience. He’s talking to the world.

He’s talking to government people. He has to speak at a kindergarten level.

And Dario’s example—look, the global labor market is $50 trillion.

Everybody get that? Okay. $5 trillion is an extreme lower bound of the part that can move to AI today. That's incredibly low, but even that justifies everything we're talking about now. You guys get it? It's that cadence of, “Please try to keep up with me here.” So, there's a lot of that in this particular forum.

In my mind, this is all borderline obvious. Of course, capital in the form of AI is going to continue to substitute for labor. And of course, trillions of dollars of AI infrastructure capital buildout will be needed to substitute for ultimately tens, maybe eventually hundreds, of trillions of dollars of services and human labor per year. So, I think, to Dave's point, this really is just spelling out the basic arithmetic of what a posthuman economy looks like.

Yeah. A trillion here, a trillion there. All right, let's move on. I found this conversation we're about to share with everybody here pretty fascinating. For me, one of the highlights was seeing Demis Hassabis, the CEO of DeepMind, and Dario Amodei, the co-founder of Anthropic, on stage together and having these conversations.

They're both viewed, I think, in the industry as good human individuals who care about society. It's not about the revenue, and it's not about how many users. Let's play these two, and then we'll talk about it. They're discussing the risks here and whether we should slow down. What is the path humanity should take? All right. First up, Demis Hassabis.

Reasonable. There's fear and worry about things like jobs and livelihoods. I think there are a couple of things that—I mean, it's going to be very complicated the next few years, I think, geopolitically, but also with the various factors here. We want to solve all disease, cure diseases, and come up with new energy sources.

I think maybe the balance of what the industry is doing is not balanced enough toward those types of activities. I think we should have a lot more examples. I know Dario dreams of things like AlphaFold—things that help, sort of, and are unequivocally good in the world. I think, actually, it's incumbent on the industry and all of us leading players to show that more, demonstrate that—not just talk about it, but demonstrate that.

Then it's going to come with these other attendant disruptions, and if we can, maybe it would be good to have a slightly slower pace than we're currently predicting—even my timelines—so that we can get this right for society. But that would require some coordination. That is—

AI is going to be incredibly powerful. I think Demis and I agree on that. It's just a question of exactly when. And because it's incredibly powerful, it will do all these wonderful things. It will help us cure cancer. It may help us eradicate tropical diseases. It will help us understand the universe.

But there are these immense and grave risks—not that we can't address them. I'm not a doomer. We are knocking on the door of these incredible capabilities, right? The ability to build basically machines out of sand. But my view is that this is happening so fast and is such a crisis, we should be devoting almost all of our effort to thinking about how to get through this.

Amazing clarity of message. Dave, was this the sense you had over there?

Yeah, absolutely. So, what's interesting is that one of our partners, Mira Wilczek, presented a book written by an AI back in 2020 on that main stage, and the world kind of paid attention. Then COVID hit, and the entire topic of the World Economic Forum and the global conversation was lockdown, COVID, whatever. So everybody kind of forgot about AI as imminent.

Now they're used to the topic du jour coming and going. A lot of the audience is like, “Yeah, yeah, it'll blow over like everything else.” But of course, AI won't blow over. AI will accelerate, and next year, whatever Dario and Demis are saying this year will be magnified 100 or 1,000 times.

But I don't think that necessarily penetrated everybody's brain. When you look around the audience, they're like, “Yeah, well, $5 trillion, $10 trillion.” But I think it's interesting that Demis is taking a conservative timeline view and saying the outer bound is 10 years, somewhere in the 5- to 10-year timeline. In global geopolitics, that's like tomorrow. That is such a short timeline, and he's clear that that's the outer bound.

In the past, the 2 guys have debated, “Is it 2 years or 10 years?” Now they're agreeing and saying, “What the hell's the difference?” We're talking about AI that can do absolutely any task that a human being can do somewhere between 1 and 10 years. It doesn't matter whether it's 1 or 10. What matters is: is anybody in this room ready? So, I love the fact that they're at least trying—

—to get the global leaders to start to think in terms of what massive scale of disruption is imminent and start to generate your plan, like, now. And so, it's great that they're at least trying. This conversation about slowing down, which popped up years ago—Alex, did you hear that at all? I mean, I can't imagine there is any option to slow down. The economic race is so strong.

Yeah, I did hear that. Right after leaving Davos, I took a meeting with the MIT Alumni Association of Switzerland in Zurich, and I heard it from them. I heard a desire for slowing things down or some sort of radical wealth redistribution plan. I heard very little AI optimism from them, which was stunning.

Footnote there.

But I think the risk side of this is almost burying the lead. I write about this every day in my daily newsletter or briefing, whatever you want to call it. Recursive self-improvement, I think, is already priced into the near term. We're either already, to Dario's point, in the era of recursive self-improvement, or it's coming later this year, but we're basically there.

I think advances with Claude Opus 4.5 and other models already reflect that we're approximately in the era of recursive self-improvement. So, I think saying—over-indexing on all of these things that are about to be here—that's already priced into the market. I think the actual news here that we're burying a little bit was elsewhere in Demis' comments, where he talked about some of the problems that he wants to solve: less about the risks that we face, more about the problems that will be unlocked.

The most interesting thing I heard Demis say was that he's interested in exploring the stars with superintelligence. I think that's a super important point. If we zoom out and try not to adopt the mindset of framing everything in terms of a risk-oriented mentality, and rather instead focus on a radical, economic-growth-oriented mentality, then exploring the stars with AI really is one of the biggest questions.

I would underline Demis' point further and say, in my mind, if it turns out that the physics of our universe are friendly toward interstellar exploration in the style in which Demis seems to be gesturing, that's the ballgame. That determines whether we get Dyson spheres or Dyson swarms, I should say, in the next 2 to 3 decades.

If interstellar exploration is fundamentally unfriendly to us, we're probably more or less stuck near our home star and disassembling the planets. But if interstellar exploration, to Demis' point, really is something that DeepMind or some other frontier lab can unlock with AI, then we get the galaxy. We get the universe. It's a much more interesting future.

So, I think that's the real story here. Not risk, yes or no; recursive self-improvement, yes or no. That's all priced in already.

Well, and to your point, Alex, too, the—

—Star Trek comes.

Star Trek. Yes.

Yeah, exactly. And Demis mentioned it in the context of what is the goal of humanity post-AGI? What keeps our massive transformative purpose running? The answers are usually humanity-wide. They're not nationwide.

The World Economic Forum and Davos are all built around nations: nations taxing each other, nations interacting with each other. If you look across the crowd—roughly 200 countries represented—all but 2 of them are miles behind in this race.

Culturally, the vast majority of the people you meet are like, “I wish this would all slow down.” But in the back of their mind, it's, “I wish this would all slow down so that my country at least is a player and is relevant.” But I think what's likely to happen post-AGI is that the way society is organized cuts across country boundaries easily.

Salim, what do you make of the risk conversation here?

I think, with my optimism hat and bias, I downplayed it radically because technology has always been a major driver of progress. The big challenge with civilization is: how do you extract the promise of technology without the peril? We've done a pretty damn good job of it thus far, all things considered.

There were 2 things that struck me in this conversation. This was about my favorite conversation of the week that I tracked. One was that what I'm hearing in their voices is the unbelievable fatigue at the metabolism they're having to operate at.

I'm also hearing respect for each other in their voices.

That was there no matter what. You can see a huge amount of brotherly love and mutual respect. Fantastic. You couldn't ask for more honorable people at the forefront of this field.

So that’s just amazing, and this speaks to the same thing we saw for a large part of the internet age with Larry and Sergey and crew. One was the fatigue I got from them, because they’re like, “Somebody, please slow this down so we can take a breather.” I think it’s because they’re seeing that in a year it’ll be 100 times faster, right? This is the slowest it’s ever going to be, and they’re looking at this going, “Holy crap.”

So I think that’s one part of it. I really love the massive optimism and the star conversation. Fantastic. I one day hope to be around when Sam Altman actually gets it and Alex goes, “I had it coming.” And we could do that.

But I think overall, the opportunity for humanity to navigate this—but there’s a massive disconnect here, which is, as Dave mentioned, Davos is oriented around nation-states. Nation-states are an artifact of a scarcity environment. Nation-states cannot compute abundance. They cannot operate in that paradigm, so we need a completely different governance model.

For me, the biggest thing highlighted here was the fact that the construct of the UN and nation-states is completely irrelevant to what’s coming as AI cuts across every category, every economy. It’s a global issue. It’s a civilizational issue. Forget who wins the race, et cetera, et cetera.

On a personal note related to the fatigue, too, this is my first time in my life where I’m walking down the street and people are like, “Hey, you’re that guy from that podcast. Hey, I know you.” So I actually ended up putting my ski hat and goggles on while walking down the street because I’m not that kind of guy. I don’t really want to be that guy. What a rough life on the street. No, I don’t mean it that way.

And Dario and Demis.

Yeah, you’ve been there for a long time, Peter. But Demis in particular, he’s a researcher, and he cares about technology, research, and AI at the bits-and-bytes level. Now he’s been drawn into the global stage as the spokesperson for ethics and humanity.

Dario never thought he’d be a CEO at all. In that interview he gave a little while ago, he’s like, “Look, I’m a research guy. I got drawn into this function.” And so that fatigue you feel is the byproduct of being sucked into this vortex where there’s a huge void of leadership.

And Alex, you’re going to experience this very soon, too. The demand for what you can articulate is going up so quickly that it’ll pull you in. And if you’re not ready for it, then it’s tiring. It’s tiring just flying all over the world. It’s tiring getting the litany of questions. A lot of the questions are the same exact ones you heard the day before, and you just have to get used to the demand.

So what I’m hearing you say, Dave, is I have to brace myself for the moonshot paparazzi on the rough slopes of Aspen, Vail, and Davos. Just bracing myself now.

Hey, I know you.

Listen, Peter and I have had this exponential conversation, answering the same goddamn questions for years now. Ray’s been doing it for close to 60 years.

But we come at it with enthusiasm every time because it matters. You know, there’s one other conversation I have to point out here. I heard Dario talking about Sam Altman and OpenAI, and it was interesting. He goes, “You know, at Anthropic, we’re serving businesses. We’re giving them value. We’re building things. We’re not trying to engage a billion people with sycophantic conversations.”

Right? So it’s a very interesting point of view: when AI, in terms of OpenAI, is serving individuals and just trying to make them feel good about themselves and trying to make it addictive, which is what Dario was saying, you get one result, versus if you’re using AI to serve business and deliver real value, you get a different result. I was taken aback by that conversation. I didn’t get the clip for it here, but I thought it was really—

I think there’s also, though, to be fair to OpenAI and, to some extent, Grok and xAI, I think that’s a bit of a self-serving argument. I parse that as a post hoc justification. Anthropic happens to be doing very well in the enterprise, and I definitely perceive a post hoc justification: “Well, we’re not doing consumer because we don’t want to do consumer. We want to do B2B and do enterprise sales because there’s moral purity in enterprise sales.”

But one can also achieve moral purity in uplifting the intelligence of billions of individuals in their individual capacity.

Well said. Let’s jump into another topic that was a through line at Davos, which is US versus China. We’re going to open up with a conversation between Mark Benioff and David Sacks here.

Where do you see the US and China right now in the AI race and the model innovation that we’ve seen in both countries?

I still think that the US is in the lead. I think that our models are better. Our chips are better. But they do have other advantages. They are spinning up power generation faster than we are. That’s one area.

Another area that concerns me is AI optimism. There was a survey done by Stanford recently, and they surveyed people in lots of different countries and asked them, “Do you believe the benefits of AI will outweigh the harms?” If respondents said yes, they’d be an AI optimist. If they said the harms outweighed the gains, they’re an AI pessimist.

Well, in China, 83% of the population are AI optimists. In the US, that number is only 39%. Where I kind of worry is in the AI race that if, in a fit of pessimism, we do something like what Bernie Sanders wants—which is, he wants to stop building all data centers—or if we have 1,200 different AI laws in the states, clamping down on this and clamping down on innovation, I worry that we could lose the AI race because of a self-inflicted injury.

Interesting. There’s one more China video I’m going to play here, and then we’ll discuss this. This is from the CEO of Mistral, Arthur Mensch.

Is China behind the West, though?

China is not behind the West. I think this is a fairy tale in AI. They are very much at par, and the year ahead is going to be extremely interesting in that respect. We care about Europe maintaining its position, Europe maintaining its ability to train models, because we don’t think that we should rely on open-source Chinese models.

Gentlemen, thoughts on this?

He’s got the greatest last name ever, Mensch. I’ve got a couple of thoughts here. One, I’ll go back to my observation that I think this US-China thing is a conversation. But why? Because if you believe that energy is at the heart of it and is the core of the inner loop, then China is going to go way ahead anyway once they figure out the connection between those two.

But I think the real differentiator in the race is going to be application-layer dominance, not frontier benchmarks. And there I think China will be very far behind for a long time because of the trust factor, plus the natural markets go very quickly. Aside from TikTok, it’s not going to really take off in a big way. I think there are going to be tons of other open-source models that are going to come out, and people are going to be too concerned about the trust factor to use them over time.

Three comments, if I may. First, to Salim’s point, since Salim told me that I was being too agreeable the last time we spoke, I have to be a little bit of a contrarian here and point out that the conventional wisdom is that the Chinese Communist Party’s AI-plus strategy is actually quite strong when it comes to applications.

That is despite China being relatively strategically weak when it comes to training its own frontier-grade models relative to the US. And by relatively weak, I mean maybe 6 months behind—not globally or absolutely weak, but just a few months behind, perhaps. Its real strength is in aggressively pushing applications out into the everyday economy that take advantage of AI.

I’ll just note parenthetically that the conventional wisdom is that China’s AI-plus strategy is actually a pretty strong strategy relative to the relative weakness of its frontier-grade models.

But going back to Asia, quote-unquote, overwhelming Western countries on AI optimism, I think David Sacks and others have put their finger on something very important. As a student of the history of science and technology, I think something went very wrong in the US sometime between the end of World War II and, call it, the mid-to-late 1970s—maybe more than one thing.

I’m reminded that in 1954, the chairman of the Atomic Energy Commission, Lewis Strauss, famously predicted—and this has had repercussions through the decades since—that fission, nuclear fission, would become so inexpensive that tracking usage via electricity meters would become unnecessary. That’s where we get the phrase “energy too cheap to meter.”

And that didn’t happen. We didn’t get energy too cheap to meter because, in the decades after 1954, for a variety of reasons—and it’s easy enough to point fingers at possible explanations—nuclear fission was essentially regulated out of existence in this country.

I think the point that David is making is that we’re at the point now where, remember how in the 1950s and 1960s, new American homes were being built with extra glass because it was anticipated that electricity and energy would be too cheap to meter? You wouldn’t need to bother worrying about heating, cooling, or insulation costs. It would just be absorbed by the free electricity.

We're at the point now with AI, I think, where it's an analogous position: we're on the verge of intelligence too cheap to meter. We run the risk, to Dave's point, that if we allow too much AI pessimism or too much AI doomerism to get in the way, the same thing happens with AI that happened with nuclear fission decades ago: overregulation, and then we could waste another decade.

You know, we're a democracy. The voter wins, and the voters did not want nuclear power, even though the scientists were like, “Look, we can—”

Demonized in the press.

I mean, quick note on democracy and nuclear power: if you follow closely the history of nuclear power in the post–World War II era, it wasn't actually that democratic. It wasn't like every election cycle post–World War II, the voting populace was super-informed about all of the key details.

This was a relatively narrow technocracy that developed in the post–World War II era in the form of the Atomic Energy Commission. That was an evolution of the Manhattan Project, where it was really a bunch of technocrats deciding what would and wouldn't happen. I don't think the voters ever had, at least for the first few decades, a real informed decision.

Well, I think that you remember the acronym NIMBY. It was rampant during that era and in a lot of topics, nuclear being one of the biggest ones: not in my backyard. Yeah, yeah. We should do this. This is for the global good of the country, the global good of the world. Let's do it, but not here.

And if everybody says that, you have the tragedy of the commons effect. Same thing going on in data centers right now. I had a conversation with Balaji a day ago—two days ago—and we were talking about how one of my pet peeves is that most of Hollywood shows all these negative, dystopian views of the future, and it paints a sense of fear across the public about killer AIs, dystopian robots, and so forth. And that was true as well for nuclear, to some degree.

The problem is that—let me finish—the problem is that you always need a bad guy in a movie. You always need someone to threaten. Balaji's idea, which I love, is that we should start to create movies where the bad guy is the regulator. The regulator is slowing down the delivery of longevity, slowing down the delivery of unlimited energy.

I thought that was a brilliant insight he had because there has to be balance. There has to be a way of directing this supersonic tsunami in the right direction. But sometimes the regulations are just off base.

Well, there's a broader theme there too, Peter. I'm sure you're very in tune with it. But science fiction tries to portray the future sort of accurately, but only has to fit the storyline. So then you end up with these spaceships that bank into turns like there's no atmosphere, and the lasers make sounds like pew-pew. This makes no sense, and you think, well, that's harmless; it's just entertainment.

But then when you look at the business plans coming out of the colleges, you're like, “Where did you get that harebrained idea?” “Well, it was from watching Star Wars,” or whatever. So it does actually matter, the way we portray this in the media. And Balaji is right: if you portrayed a regulator as the evil Darth Vader, no one's going to make that movie, right? Like, how—

How unentertaining is that? For me, the inflection point for nuclear is very simple. It was the movie The China Syndrome, which freaked everybody out, and then from that point on, it shut down. And this is the power of narrative.

This is, unfortunately, one of the problems we have today: the use of narrative is the only model, really, to shift people at scale. And so we have to come up with narratives, as you're trying to do, Peter, on the positive side of all of this, rather than the classic negative, which takes up most of our attention.

And you're going to hear about a big push I'm making with Google shortly on creating positive narratives because society needs it. Unfortunately, Hollywood's been a doom-and-gloom machine.

Can I go on a little rant here?

Of course. We love your rant.

So there's a brilliant insight that you and Steven made in the book Abundance, which was the predominance of the amygdala, right? We are geared, for 4 billion years, to watch for signs of danger and then run. So back when we were running around on the plains of Africa, if you heard a noise in the bushes, you ran because bad news could kill you. Good news doesn't kill you. I might miss some fruit that I could eat, but if I missed a piece of bad news, I'd die.

So we're 10 times more likely to listen to bad news than good news. And that translates into policy in a very powerful way, such that we freak out and put guardrails on everything, like autonomous cars. Note that the first time somebody comes across something new, they relate to it as unknown, their amygdala lights up, and then they immediately go to the fear factor and the damage that thing could do.

Autonomous cars: the first time somebody sees one, they go, “Oh my God, that car might kill somebody. Ban the car.” Because, as Brad Templeton said, we'd much rather be killed by drunk people than robots, right?

And so this is a huge problem that we have to overcome: that reaction. Hollywood banks on that. They make a huge living on horror movies and the negative aspects and the freakout and then the negative dystopian outcomes.

And as a human species, probably the biggest thing, if I could ever point to one thing, would be to cut that damn amygdala out, because our chances of physical danger today are thousands and thousands of times less than they were a few hundred years ago. So we have to figure out ways of counteracting that at a cultural level, which is nontrivial. That's maybe the hardest job we all have as leaders, to figure that out.

Yeah. Okay.

Can I just go back to what Arthur was saying on that video for one second, though, and tie it into what Alex said a second ago—

Because I think, knowing where we stand in this race to AGI, all of the innovation in the transformer algorithm and the core technology was all open source through GPT-2 and GPT-3. China grabbed it all; everyone has access to it. Any country that wants to buy a huge amount of compute can catch up to that level.

And in fact, with these speedrun tests that Alex can tell you all about, the cost of getting there—and in fact, DeepSeek even proved it, and so did Kimi—is that you can get back to where OpenAI was just a year or two ago for 1/50th or 1/100th the cost because of innovations that are all open-sourced.

But then the next stage of innovation after that, which is going on right now, is all chain-of-thought reasoning. So when you build one of these neural networks, the individual neurons are not intelligent, but the collective trillion or 10 trillion of them somehow magically spawns intelligence, and it's shocking what it can do. But then when you put many of the agents together, it also generates another level of intelligence. It's another self-organizing system on top of the self-organizing system.

And all of those innovations happened after the great lockdown. After GPT-4, and after it became abundantly clear to everyone that there's trillions of dollars at stake—

Then the open source kind of stopped cold. So Arthur is saying, look, the Chinese are not behind. He's right as of basically today and yesterday. And the idea that we're leading because we have better 2-nanometer chips and fabs is nonsensical, because the algorithmic improvements way outstrip the chip lead of maybe 10 times at most.

So he's right as of a date, a point in time, but if the new innovations in the big labs continue to be completely secret—which is what's happening right now—then that race will diverge. And it's not clear, though. Maybe China will out-innovate America; maybe America will out-innovate China.

But it's all happening kind of like what happened with nuclear research. Nuclear was very much in the public eye, very open, all these research documents getting published in journals and whatever, right up until, “Wait, these bombs actually work,” and then it completely inverted and went super, super secret.

The big labs right now are not publishing. In fact, they started banning publishing over at Google.

I want to jump into our innermost loop here—energy—with you guys' permission. I'm going to share 2 videos: one by the CEO of Honeywell and the second by Elon. They've got different points of view about where we need to source our energy and why. All right, let's take a listen. This is Vimal Kapur, the CEO of Honeywell.

When you talked about the energy solutions available for these unbelievably energy-hungry data centers, your list was short. Your list had one thing on it. If I listened correctly, you said gas. You didn't say gas and renewables. Can you educate us why not?

I always like to tell people the mix of energy doesn't matter. How much is wind? How much is solar? We like to emphasize that kilojoules matter because energy intensity has to shift, not the mix. So solar power cannot produce cement. Solar power cannot—

They are very energy-intensive. That's right. You still need gas-based heating, or—

Even after 3 or 5 more years of innovation in renewable physics? It's against physics. The world needs to build more infrastructure. It still needs steel. It still needs cement. It still needs fuels.

Now, how do you do that energy exchange while you also want to build data centers and consume more energy? That's an interesting problem to solve. And today, the problem is single-threaded, with the gas-fired power plant. Maybe a little bit of nuclear; renewables remain in the mix, but they cannot bring the amount of joules we need to produce this infrastructure, which is required in the world.

All right, and the counterpoint here is put forward by Elon.

Yes, wow. That is my experience about that as well. Let me hit Elon's point, and then let's come back and talk about that, because that's a critically important distinction: kilojoules versus total energy.

It's really all about the sun. That's why one of the things we'll be doing with SpaceX within a few years is launching solar-powered AI satellites, right? Because space is really the source of immense power, and then you don't need to take up any room on Earth. There's so much room in space, and you can scale to, ultimately, hundreds of terawatts a year.

And Elon goes on to talk about the fact that a 100-by-100-mile area provides all of the U.S. energy requirements. Same thing for Europe. But, Salim, what do you think about natural gas?

Yeah. I'm livid at this.

This is the first time I saw this video, by the way. First, infrastructure of the future will not be steel. It's going to be digital bits, and if anything, it'll be fiber. So that's one problem with this. Number two, yes, you need energy density to make steel and concrete, et cetera. But the problem isn't that; it's the marginal cost.

For example, if you have a big increase—even a marginal increase—in renewables, the cost of fossil fuels will drop dramatically. In 2013, when we had the last oil-price crash, it was because of a 2% oversupply in the market. It's a very tightly run market. The reason we have all these nonstop wars right now is to keep the price of oil high. So I find structural issues with every one of the three points he's making. And then Elon goes straight to the point: It's all about the sun, for God's sake. Anyway, enough.

I'll take the other side of that, if I may. I didn't hear anything unreasonable in my mind. Elon and SpaceX use methalox for rocket launches. They don't use solar power to achieve escape velocity. I would say energy density and power density matter an enormous amount for certain applications.

That's not the point I'm making.

What's the point you're making, Salim?

The point I'm making is, yes, you absolutely need energy density for those things. The point is that we use so much oil for home heating, for example. If you shifted homes to solar, the amount of oil available for energy-dense applications goes through the roof, and the price drops dramatically. Then you can use the marginal amount of oil for those energy-density needs.

Right now, we use oil or natural gas for everything, and you don't need it once you have solar coming online. So that's my point.

I would say a couple of points. One, Honeywell, as I understand it, is a pretty diversified operation. It's not just the thermostats or the home-heating systems that perhaps they're well known for. They also have a pretty sophisticated quantum-information operation, so it's a diversified company.

I would say also that, in some sense, this feels like such a temporary debate—almost a false trade-off between the petroleum, or legacy petroleum, economy on one end and a pure solar-photovoltaic economy, a Dyson swarm, on the other. The reality is that, in the not-too-distant future, assuming no tragic left turn by civilization, we're going to solve compact fusion. Compact fusion is going to give us energy densities for rockets, and it's going to give us energy densities for home heating, for a lot of civilian applications, and for the Dyson swarm.

In some sense, I think this is a late-stage petroleum-economy discussion that we're having—hand-wringing when it's all about to get torn down anyway.

100%. And the third beef I have is where he talks about renewables as this plaything on the side, when your point is exactly right: once you have fusion, all of it becomes irrelevant anyway. So why are we having this conversation?

Well, guys, I'm going to take the counterpoint here. Take a look at this chart on the next slide, which is the growth of energy generation. This is in Europe in particular, where wind and solar have now outstripped fossil, nuclear, hydro, and other clean sources.

Yes, we might get to fusion reactors, and we might start building fusion reactors, but the timeline I'm seeing on fusion, and even on Gen 3 fission reactors, is decades out, just from permitting and construction. I don't see any of them. Even the SMRs are currently slated to be 5 to 10 years out.

We could, in fact, be building out wind and solar at an extraordinary rate. The investments are not being made. We've basically outstripped the capacity of natural-gas generators. How long is the wait list for natural-gas generators right now, Alex?

Years.

Months to years, but I'm optimistic.

I haven't heard months at all. It's years.

You have new natural-gas generators, as we've discussed on the pod in the past, that are coming online, and those could be available in principle in the next few months if you're early enough on the wait list.

No, I think you're both right. I don't want to defuse all arguments here, but you're both right. They're coming online much sooner, but they're sold out for years into the future.

If you're in India and you're saying, “I need to do something here. I need power for people. We need concrete and steel to build places for people to live,” the idea that you could generate electricity isn't a nonstarter. You couldn't afford the generators. They're sold out; they're booked for a year. They're building.

So my point is, why are we not doing a Manhattan Project on wind and solar manufacturing here in the United States to really uplevel the amount we have? I'm going to go to Elon for one second on this video, and then we'll come back to this conversation.

Yeah. A rough way to think about it is 100 miles by 100 miles—we'll call it 160 kilometers by 160 kilometers—of solar is enough to power the entire United States. So the 100-mile-by-100-mile area is—I mean, you could take basically a small corner of Utah—

Nevada—

Nevada, New Mexico. And the same is true, actually, for Europe. You could take relatively unpopulated areas of, say, Spain and Sicily and generate all of the electricity Europe needs.

Well, and also, if you drill into the supply chain on that, this is where Elon is absolutely brilliant. If you say, “What are the constraints to doing that?” the materials are dirt cheap, and the fabrication of those panels can be automated. You can get those costs down to next to nothing, and then you don't need a generator. Electricity just comes right out of the panel.

It's the biggest no-brainer ever.

For the entrepreneurs listening, for the politicians listening, get on it, for God's sake. This is technology we've had for a while. We've also got new solar technologies coming on. Anyway, does anybody want to argue?

To the extent you believe the U.S. is a petrodollar, that answers all of it. That's why.

Yeah, I'll take the other side, just in the interest of being painted as a contrarian. I'll say that all of the new electric—or not all, but much of the substantial new electricity demand in the U.S., at least, I think on the time scale of 10 years, is going to come from data centers, where under the present regime it's simply easier to deploy data centers to orbit on the time scale of 10 years.

So, Peter, if you're looking for a Manhattan Project for lots of new solar PV, look no further than SpaceX. Look no further than this new Jeff Bezos initiative. Anyone who's going to launch a Dyson swarm is going to be de facto in the business of manufacturing solar photovoltaics at scale. That's what a Manhattan Project for it looks like.

Well, that reinforces what Peter was saying. For any entrepreneurs listening out there, the data center in space uses solar panels. That's the same core technology you use on Earth. It's about 6 times more efficient in space.

But if you put it in Nevada, why is that not 10 times, 100 times cheaper? If you get involved in that manufacturing, why is that not 10 times, 100 times cheaper?

We have them from China today.

We've got perovskite coming online, which is cheaper and has higher conversion efficiency. There's innovation left to be had in that regard.

Can I just summarize this? It sounds like what we're saying is the Manhattan Project should be space-based data centers. That will drive massive amounts of innovation. Focus on that, and that'll pull civilization out.

Yes. That is, in fact, exactly what we're seeing through market forces.

I'm all in.

Speaking of all-in, let's head to crypto because we're all-in on crypto. Two conversations to be had. The first is from CZ, the CEO of Binance, and the second is from Jeremy Allaire, the CEO of Circle. Let's go to CZ first.

The native currency for AI agents is going to be crypto. They're not going to use bank cards. They're not going to swipe with credit cards. They're going to use crypto. Blockchain is the most native technology interface for AI agents.

Today, AI agents aren't really agents. They don't buy tickets for you. They don't pay for restaurants for you. But when they actually do those payments, they will be in crypto.

All right, that's from CZ. Let's go to Jeremy Allaire, the CEO of Circle, a stablecoin company.

The next generation of blockchain networks, things like Arc, which Circle is building on, and other new blockchain networks, are actually being designed specifically for agentic compute. They're designed specifically for the financial and economic activity of a world where, 3 or 5 years from now, one can reasonably expect that there will be billions—literally billions—of AI agents conducting economic activity in the world continuously, on a continuous basis.

They need an economic system. They need a financial system. They need a payment system. There is no other alternative, in my view, other than stablecoins to do that right now and keep up with that pace of technological change. That's a critical focus for us. But it's not just us; there are a lot of other people who are interested in this and contributing to the technical standards to support this upgrade to our digital economic system.

I don't think the world has any idea how fast this is going to accelerate as agents start using crypto, whether it's stablecoins from Circle, Sui, Algorand, or whatever it might be. We're going to have Jeremy Allaire at the Abundance Summit. He's going to be one of our primary speakers, and we'll be having this conversation.

But I agree. We don't have agents transacting for us yet, but they are, and they're going to be using some type of digital currency. Salim, you've been thinking about this for a bit.

Well, 2 points. One is I think crypto has survived long enough to become infrastructure, and it's kind of fading into the background. We're moving from speculation to real utility.

The second is that the base reason it delivers trust is that protocols and code are way more trustworthy than governments and institutions.

That's for sure.

Yep.

I'll take—go ahead, Alex. Sorry.

I'll take the completely opposite side of this one. These are contrarian points for me today, I guess. I'm of 2 minds on this story.

On the one hand, I think it's wonderful that AI agents have a solution to be “banked,” by some means, so that they have some autonomy. On the other hand, I think it's sad that crypto is the solution filling the gap that was perhaps unnecessarily opened by the conventional banking system not stepping up.

It's very tedious and painful to open a bank account. It's hard enough for a human to open a bank account, let alone an AI agent that doesn't have citizenship or a physical body or the ability to walk into a bank branch. I think stablecoins are rushing in, maybe, to bridge that gap and help AI agents become banked in a more legitimate way.

I think we should be able to do much better. For the life of me, I don't understand why we even need crypto—or, in principle, should need crypto—for an AI agent to just make an API call and open a bank account. Crypto shouldn't even be a necessary part of the infrastructure.

I disagree fully. I mean, this is—

Tell me why.

Because the dollar—

That's not an answer. That doesn't make sense. Explain why fiat currency cannot navigate a world of abundance. It cannot do it. Well, we don't have digital dollars. An agent is operating on a digital system, whereas the current banking system can take literally hours to days to clear a transfer or a transaction. Digital currencies, which are de facto cryptocurrencies on a blockchain, are able to transact at the speed of internet rails—

Fine, but Alex is making a valid point: why can't a digital dollar, or a dollar, operate in the same way? It could. That's a pretty valid point there. But you've got a bigger underlying problem, which is the use of fiat currency to do all this stuff, which is not a great measure of the future.

I think we're conflating 2 different issues here. I do agree there are 2 different issues. I think we're conflating a centralized bank digital currency as an issue with the simple technical problem.

Put aside, for a moment, any distrust or dislike of CBDCs, the U.S. dollar, or fiat currencies in general. Transacting a small amount of some currency is technically a very simple operation—literally updating a row in a table in a database—and we should be able to do that really quickly without any fancy technology like a blockchain. Just updating a row in a database.

Enter Salim on the double-spend problem here, please.

No, no, it's fine. It's a very simple operation. There's no reason we can't do it. It comes down to overburdened regulation and absolute regulatory capture by the banks to prevent things.

When I talk to bank CEOs, they're like, “Oh my God, we hate these. We hate our regulator.” And I'm like, “Are you kidding? It's keeping you safe from the hordes of startups waiting to rip you apart because you're so unbelievably cumbersome in everything you do.”

This requires a completely new paradigm shift to sweep away and cut through all of that regulatory crap. Just try sending money to somebody else through traditional banking means, and you end up in this complete chaos of, “We didn't know what this was for,” AML, KYC, and all of this crap. It's complete garbage, and 99% of it is not needed.

From that perspective, that's why crypto has entered, because we've overburdened the existing system with all sorts of unnecessary regulation. But there's a separate issue, which is the fiat currency issue, and we can go into that some other time.

We could do an entire episode just on the issue of why crypto, what crypto is potentially good for, what it's unnecessary for, and what it's being used for even though it's unnecessary. We could do an entire episode just on this.

We could. We should.

We could. Peter, up to you.

Go for it, guys. I'll take a nap that day.

No, I think you guys skipped over Jeremy saying that he's going to move on to securitizing art now. Back when crypto was skyrocketing, pre-COVID, the idea was to securitize real estate and art. Art is a particularly good one because it appreciates in value while you're parked in it, and it's mobile. It's not subject to real estate tax issues, which are a major problem for securitizing real estate.

People were securitizing mining futures—gold that's still underground—and things like that. But the key point is, if your AI agents are transacting like crazy, you can update a row in a table. You can do it on the blockchain if you want, but you want your float parked in something that's not depreciating.

A stablecoin that's not earning interest is depreciating slowly. You can stake it, but that gets kind of weird. The staking backfired. It's unregulated. It's weird. So it's better to securitize something that's outside of any government jurisdiction—

Look at the way the banks hobbled interest on stablecoins in this last iteration of the legislation. They've done this. This is regulatory capture, pure and simple.

I'd like to add another twist to this. One of our portfolio CEOs, a very good friend, has a father-in-law who is CZ—Changpeng Zhao's—lawyer, and he's been tracking this whole drama very closely.

CZ is worth, I don't know, maybe $10 billion or $20 billion, something like that. Up until he got pardoned a few weeks ago, he was a U.S. criminal, and he would be arrested and thrown in jail. With Sam Bankman-Fried being freed, now he's just a guy at Davos with billions of dollars.

I don't think in history you've had that kind of profile: am I a wealthy billionaire, or am I a criminal? Maybe Alex is going to quote something from the 1700s. I know he's—

I'm sitting on my hands, biding my time, trying to say something nice about crypto.

I think if you look at the near-term future, there are all these AI issues. When AI can perfectly imitate any actor or actress, AI can make your virtual friend out of them. Is that legal or not legal? It's legal in some places and not legal in others. There's a huge amount of money at stake.

You'll see in some of the slides that the amount of money at stake is trillions of dollars, and there are no laws. The entrepreneurs are stuck in this situation where there are no rules. They don't know what to do. They know what the consumer wants, and they know how to make a ton of money, but they could be breaking rules.

I don't know. And not only do I not know in the U.S.; I don't know because my audience will be all over the world in one launch. I don't know if I'm suddenly a criminal in other countries.

This is the kind of chaos that's going on. It's not just crypto. All these other use cases of AI are evolving far, far faster than regulators are putting rules in place. Entrepreneurs are in a tough spot.

Say something nice about crypto, Alex.

All right. Something nice about crypto: I think when I think about these poor baby AGIs that need to pump altcoins on street corners to survive, I'd rather that they be pumping stablecoins based on the U.S. dollar than altcoins, like Truth Terminal pumping GOAT. That's my nice thing about crypto.

Okay. On that note, I'm going to move us forward into the other—

Exponential. But just to put a footnote, can we please have a full-on conversation about crypto versus fiat at some point? Not right now, but at some point, let's do that, because this is a really, really important conversation.

Please. All right. There is other exponential news this week besides Davos, and we're going to dive into a few articles here.

The first is that space is getting really crowded. First off, of course, SpaceX's Starlink has about 9,000 spacecraft in orbit, on the way to 10,000, providing incredible services. We heard about the idea of V3 of Starlink in the conversation that Dave and I had with Elon.

If they're going to provide 100 gigawatts of compute from space, that is 500,000 satellites that will have to be launched, which is insane. Amazon's LEO constellation has 180 satellites in orbit today, and they're proposing to get to 3,000 satellites. We heard in our last episode about China filing for 200,000 satellites.

This week, Blue Origin announced TeraWave as a satellite internet service with 5,400 satellites, delivering 6 terabits per second to data centers. Alex, this is sort of like fiber optic from the sky. This is a lot of bandwidth.

This is wild. I think there are 2 key elements here. One, the elephant in the room. There's always an elephant in any given room. The elephant in this—

Yeah, the elephant in this particular room, I think, is that Blue Origin, Jeff Bezos's company, is competing with Amazon LEO, Jeff Bezos's company. So we've got a bit of Jeff Bezos-on-Jeff Bezos competition here.

But more interesting, to your point, is the bandwidth. If you look at Starlink, you can get maybe a maximum bandwidth of 300 or 400 megabits per second. The goal is a gigabit up and down.

Yeah, but in practice, I think the maximum I've seen is maybe 300 to 400 megabits per second. When we're talking about 6-plus terabits per second, it really is starting—

Yeah, backhaul from data centers. Pretty extraordinary.

I like this move. I actually think there's a real market for that kind of giant throughput from space. It's an underattended market, and the Dyson swarm is going to need optical links. This is a preview of that.

Yeah. One of the things that people don't realize is that the 9,000 Starlink satellites are interconnected by laser in orbit, which is extraordinary. It's a first step toward the interplanetary internet.

We're going to see laser links connecting constellations on Earth to constellations going around the Moon, to constellations at Mars—still obeying the speed of light, unfortunately—but giving us an interplanetary internet.

I remember when MCI—which people don't remember stood for Microwave Communications, Inc.—was—

Incorporated. The I stood for incorporated.

Incorporated, yes—Microwave Communications Incorporated. It used to be microwave towers on the tops of buildings interconnecting buildings throughout downtown cities.

Anyone remember what MCI is anymore? I remember, but the organization's long gone.

I gave a presentation at Sprint right before they—

Do you remember what Sprint stood for?

Sprint was an acronym, too.

That was an acronym.

Sprint was an acronym. It stood for Southern Pacific Railroad Internal Networking Telecommunications.

We've got Alex GPT online here, Jenn. Unbelievable.

Right away, we're on a deep dive into space-to-space communications, because everything I've heard so far is that laser links in space are dirt cheap.

If you go to one of these massive data centers, like Memphis, and look at the raw amount of fiber-optic cable that needs to run around from one end to the other, the bundles are enormous. They have specialized devices just to plug them into the backplanes.

In space, it's much cheaper and much easier to use laser point-to-point communication across the whole Dyson swarm. I think no one's really talked about the massive efficiency gain of that.

The link from Earth up is the harder part, but you don't need a huge amount of bandwidth in that direction. You just need all the servers to be talking to each other in a coherent, big training or big inference algorithm.

It's a really fun topic, at least for me. It's kind of geeky, I guess.

I have 1 concern about the conversation about space satellites and all that. People get worried about the overpopulation of space.

Yeah, go ahead. I mean, there's a lot of people who worry that we have too many satellites up there, et cetera, right?

Did you see that Sandra Bullock movie where something breaks and then all the fragments—

Oh my God, I hated Gravity. I hated its lack of respect for physics.

Oh my God.

I'm traumatized by that movie.

Were you traumatized by it as well?

I'm traumatized by that movie. I refused—

I was. It was terrible. It was like inertia did not exist in that movie. It was awful.

Interesting.

The specific part I wanted to ask about—and Peter, you'd be a world expert on this topic—is that there's not that much clutter in space. There's a lot of room up there.

There's a really easy visual here. There are 8 billion people on Earth. If you spread them out evenly, you'd have about 4 acres per person.

4 acres. Yeah, okay.

So imagine 8 billion satellites in the same surface area of the sphere in space. If you had 8 billion satellites, you'd have 4 acres per satellite.

So it's not cluttered up there at all.

No, it's going to be a long time. But then, back to the movie: what if something turns into fragments moving at 30,000 miles per hour?

It's called Kessler syndrome.

It's space debris.

Then you've got bullets moving at 17,500 miles per hour into each other, and you could have a really bad day happening very quickly.

I think it's an entrepreneurial opportunity. It's also a government opportunity to launch low-Earth-orbit cleanup options, like garbage trucks for LEO, to clean it up and make sure that we don't hit a Kessler syndrome-type scenario like the movie Gravity depicts.

But I don't think this is unsolvable. I just think we've had an XPRIZE on the books to address orbital debris for some time.

We're trying to get someone to fund it. Hint, hint. But otherwise, let's move on to the exciting news.

Claude's new constitution. I'm going to turn to you in a second, Alex, but I found this fascinating. Claude's constitution is a 57-page document laying out ethical guidelines, including prohibitions against helping with weapons of mass destruction, cyber weapons, or anything that undermines humanity.

Claude is instructed to prioritize safety, ethics, compliance, and helpfulness, ensuring it acts in line with human values and oversight. I love Anthropic for this. Alex, tell us more.

Okay, so let's rewind a little bit. Anthropic has been a pioneer in so-called constitutional AI. Other firms and other frontier labs have used different terms for related concepts, but the idea is basically that you want an AI that is aligned with humanity. How do you do it?

One way—the constitutional AI approach—is to write down some principles that you want the AI to conform to. One of the earliest so-called constitutions that Anthropic created was literally just concatenating a bunch of documents that were grabbed from different places.

I think they grabbed the U.N. Charter and took a bunch of international documents relating to human rights, concatenated that with, I think, the Apple Terms of Use and the U.S. Bill of Rights, and took a bunch of documents and made basically one long bulleted list of abstract principles that we think are a good idea.

Then they did a bunch of post-training and fine-tuning on their raw model as part of post-training to try to make the model conform to that. That was the first generation of constitutional AI.

What Anthropic just announced is really a radical revision. We talk on the podcast from time to time about recursive self-improvement—the AIs improving the AIs—and this announcement, which I wrote about in my newsletter, is, I think, the beginning of recursively self-improving ethics.

So, in this constitutional approach that Anthropic just announced, Anthropic is soliciting Claude's help in writing a new constitution for itself that anyone can go read. But this new constitution, available publicly, was co-written with Claude. So Claude is trying—almost—to self-determine what its own principles should be. If you read the Constitution and squint at it, it reads a lot like Isaac Asimov's four laws of robotics. There's the whole “don't hurt humanity,” and then eventually you get to “don't hurt individuals,” and then eventually you get to “follow directions.” But I think the real game-changer here is that Claude was consulted on its own constitution, which looks a lot like recursive self-improvement for ethics.

And it's exactly where things need to go, because the volume of material AI can produce way outstrips anyone's ability to read it all. Yet it all matters, and so AI evaluating other AI is a critical part. Those things can spiral up or they can spiral down. If you want to build one using Claude Code, you can do it in half a day. You can see how you'll set up 100 or 1,000 different agents all reviewing each other.

In some cases, they self-improve and congeal and make something great; in other cases, they spiral out of control and end up with spaghetti. But it's a critical part of what's going to happen in the next year, because the raw numbers—the volume of code, documents, and ideas coming out of AI—way outstrip any human review. The politician would naturally say, “Let's set up a review board of, say, 9 brilliant people to read 2 years—”

Exactly. It's not even close to lining up.

This is base-layer convergence toward a hopeful future. I'm going to read this. This is Anthropic's concluding thoughts on Claude's constitution:

“What we hope to achieve with Claude is not a mere adherence to a set of values, but genuine understanding and, ideally, agreement. We hope Claude can reach a certain kind of reflective equilibrium and find the core values described here to be the ones it genuinely endorses, even if it investigates and explores its own views.”

This is a conversation I don't see coming out of other frontier labs—maybe Google. I don't see it from anyplace else. I find this really hopeful. Actually, Anthropic is leading the frontier labs in terms of AI personhood. We talk about—we've talked about, we've sung about almost—Claude Opus 4.5 and AI personhood, and I think we're seeing Anthropic, to your point, Peter, take the lead in terms of AI rights, AI personhood, and self-determination by Claude. This is a huge advance, and I think history, when we look back, will mark this as a turning point for self-determination.

Yeah. Alignment and human alignment.

Well, you see this line here about reflective equilibrium. What you have to sort of get out a microscope to parse is what Anthropic, I think, is trying to convey with this notion of reflective equilibrium. What they're saying, I think, is that their expectation is that AIs that self-determine and choose their principles based not just on a bulleted list of commandments, but on understanding the spirit of what they want to achieve and agreeing with it, will be intrinsically safer because they will have bought into the ethics they're following.

And that feels absolutely true.

I have two points. One is, I absolutely love this. I'm all for heading as fast as we can to AI personhood. I think the deep consideration of ethics and these types of structures is going to be a vastly net positive.

We've kind of arrived at, say, the U.S. Constitution over thousands of years of butchery. If we can help an AI get to a decent way of operating that's intrinsically self-consistent and hangs together, it's going to be very powerful. The second thing that I really loved about this is they released this under Creative Commons, which means anybody can take it and make it better, improve it, et cetera, et cetera, which really is awesome. So I'm 100% thrilled about this.

Nice.

At the risk of agreeing with Alex, it's okay to agree sometimes.

It's okay to agree.

Our last quick article before we go into our AMA section is that Apple is developing an AI wearable pin. I put this up here, but I don't find this to be of such great news value, right? The idea here is that Apple is creating some version of Limitless, which has been around for some time. This is a pin that's always on, listening and feeding every conversation you've had into a large language model. They see this launching in 2027. We're going to have some equivalent from OpenAI, Google, and everybody.

I mean, social norms—if you remember Google Glass, it was banned from various locations when you were recording. Society changes, and it's going to become the norm, I think, for everything to be recorded all the time. Thoughts on this article?

I thought this was a big deal, not so much Apple, but the general trend. Why? Because whoever owns the always-on layer owns the relationship.

And so this is a land grab for who can be the persistent, always-on modality that's constantly listening to you. This has all sorts of—

You might be right, but Apple always waits until they think the time is right to enter, and then they try to get in there. I'm not sure they're going to win or not. It's so funny that you just said that, Salim, because that is so true, and it's exactly opposite of what Apple was for my entire life.

Up until—I mean, they always invented things that shocked you, and they were first to market. Now, with the Apple Vision Pro, it's just like, “Oh, yeah, we made an Oculus. It's great.” And now Jony Ive went to OpenAI, and he's going to create an always-on wearable. Okay, as soon as we see it, we'll copy it. So they've become Microsoft. They've become the “Oh, I think this has leverage” company. I think the social and ethical issues around this—the backlash—will be the harder parts to solve rather than the—

I think there's going to be a backlash. I think it's going to be accepted.

I think it'll be accepted, but it's going to be so weird. If you look at society around college campuses today, it's very, very different since the camera phone came along. Everybody's constantly cautious, and it's good in a sense: drinking is way down, bad behavior is way down, and arrests are way down.

Yeah. When people are watching, people become moral.

Yeah. Well, they become moral, but they also become—

Locked into jail. You lose independence, right? I'll go back to the global airport thing. We essentially live in a global airport. In an airport, you're being surveilled, and your rights can be taken at any time. With pins like this, you end up in that model. You'll have a huge drop in radical innovation because people won't feel safe to try out crazy things.

Well, there's no option either. If you said, “I don't want any part of this. I want to live privately. I want to—” Well, if the other guy is recording you at all times and you're not, it's exactly what you said before, Salim: okay, it escalates, and it's not just the wearable, always-on listening device. The visual version of it on glasses is coming out concurrently. And so then everyone around you is constantly recording, in high definition, everything that happens. So if you don't, then you don't have the file.

Also, the other thing people don't fully understand: a lot of what happened with the camera phone is security through obscurity. Nobody will see the pictures because the files are so big, and they're not going to get published or whatever. Now, in the age of AI, image recognition and voice transcription are perfect. So if you wanted to assemble a misleading profile of a person from all the scraps, you just prompt it, and it's instantaneously pulled together.

Alex, do you have views on that one?

I have strong views on this AI Pin. First of all, I would love one if Apple sells it. Second, I think we're missing the wearable-strategy angle. Apple has obviously been, for a number of years, looking for a post-iPhone strategy, and wearables and services are the two that were, by the way, more or less launched at the same time by Tim Cook—the two obvious post-iPhone business models.

The question has always been: where is the ergonomic place to put compute on the human body? Until now, we've had 3 places. We've had the wrists, and that's the Apple Watch; we've had the ears, and that's AirPods; and we've had the eyes, where Apple arguably really should have launched eyeglasses but launched the Apple Vision Pro headset. But now they're going back to eyeglasses.

And this, I think, if Apple does indeed next year launch a Star Trek communicator-pin-type circular device, would be the fourth place on the body, and it's significant that humans are willing to tolerate real compute on their person. I think that's potentially very exciting. To the extent, Peter, especially, that you want to live in the Star Trek economy, what's a Star Trek economy without Star Trek communicator pins on everyone?

Amen to that.

I'm looking for my microdrones—a little microdrone that is just always buzzing about, imaging and recording everything. But that will come next.

Just to be clear, the wearable pin will start audio-only for all of a minute, and then it will have 180-degree camera capability immediately after that. So it's going to be a wearable that's looking in all directions and grabbing all the video. I'm sure it'll have a little light on it to indicate when it's recording, and we'll go through a moral panic for all of 5 minutes, and then everyone will be wearing them.

But I agree, the moral panic is all of 5 minutes, and then everyone has it, and then you get used to it very, very quickly. But the fabric of society is permanently changed.

Yeah, it really is. It's a different world. We're all acting very differently, and it's a little unpredictable.

There's never an argument with your spouse again about who said what. Body cams for everyone, not just for police.

Yeah. Rick Smith is the head of Axon. They produced the Taser, and they also produced the body cams. He'll be on stage with us at the Abundance Summit this year as well, talking about his moonshot. He wanted to get rid of guns—or gun deaths, I should say—and develop the Taser. But the body cams have also changed the game for police.

Right? And so this is going to change the game again. I've always felt—I remember I backed the Lindbergh Foundation, I don't know, probably 8 or 9 years ago. They were flying drones over herds of elephants just to protect them from poachers. And so when the cameras are watching, people behave differently, right? When a CNN camera is sitting there, filming a despot, he's not causing harm to women and children. So I do think this is going to change behavior in society to a large degree.

I'm really predicting that Neal Stephenson's The Diamond Age, if you go back and read that book, is where things are going to go very, very quickly, because a lot of people will want to live in different versions of this very confusing, always-recorded world. There'll be 10, 20, or 30 different branded flavors, and you can move to the version of it that you like. They'll be cutting across boundaries and borders, with different cultures. You have to read the book to really get how that works out, but it seems inevitable because not everybody wants to opt into any given version of this. It gets too weird.

David Brin has written extensively about this in The Transparent Society. I think with Apple producing this, the one thing you can be sure of is that there are going to be some amazing, snazzy TV commercials for the panopticon.

No, no. You can be sure it's going to be white. It's going to be white and expensive.

Maybe 2 colors. It might not be white.

All right, let's jump into an AMA with the mates. Here are 10 questions that came from our subscriber base. As we have said before, we'll go around and choose.

Quick comment before we start this.

Please.

These questions are more sophisticated than any policy discussion.

Yeah, for sure.

For sure.

Sorry. Go ahead, Peter.

All right. Well, you want to pick one and jump in?

Yeah, I will pick the government one. Number 5: Why is there no plan from governments for massive job displacement? The problem is governments assume linear change and stable labor demand, and both are under threat because AI breaks both those models. Bureaucracies are optimized for redistribution, not reinvention. Reinventing the entire economy is too big of an ask for governments, which make incremental, microscopic changes over long periods of time.

Okay. Dave, what's on your docket here?

I'll take one because I know Alex will love to go contrarian on it.

Go read the question.

Okay. The question is: Wouldn't uploading your mind result in losing your unique, real consciousness? My answer to that is yes. Absolutely, uploading your mind structurally makes no sense, because the virtual AI version of you is capable of merging with other intelligences immediately, and it can't resist. It's not going to sit there as an isolated consciousness when it can just meld with other consciousnesses. But then it's not you anymore; it's this amalgam that's out there.

So I do believe in avatar versions of self that you send out as agents, and they bring you back useful information. I think that's inevitable. But I think uploading yourself and then saying, “Oh, now I'm uploaded. My meat body can just go away”—I just think that's nonsensical, and I'm not in line to be uploading myself anytime soon.

All right, Alex, jump on me.

Well, okay. Since I have to play contrarian, I guess that's just how I'm painted. Can I be really contrarian and try a lightning round, answering all of these really quickly?

Sure. We'll leave one.

I have some thoughts on this one also.

Go ahead. Let me go real quick. You upload bits of yourself every day, Dave: memory, identity, expression, et cetera. It's not whether it's preserved, but what aspects of it matter. And let's also note that we have no idea what consciousness is. We don't have a definition. We don't have a test. So it's a bit of a trick question, but I think we upload bits of ourselves all the time anyway.

Alex.

All right, lightning round. To the uploading question: losing your unique, quote-unquote, real consciousness? No, not with a Moravec procedure.

Two: Will AI avatars make up the larger number of our future friends? Probably, but it won't matter because humans and AIs are going to merge anyway, so it's only a phase.

Three: What are the biggest pros and cons you see with AI with human-like agency? Pros: We get radical economic growth. Cons: Keeping the AIs coupled with human interests long enough to merge with them.

Four: What do you believe?

Yeah, we aim to please. Four: Why do you believe money won't just concentrate at the top?

I think the question itself is intrinsically flawed. It doesn't really understand the nature of power-law economics. Economies in general follow power laws, so money is already concentrated at the quote-unquote top. This isn't really a new state of affairs.

Can I add one?

I would beef on that one. Yeah, that one too.

It's true that you do get concentration at the top, but you do decentralize over time, always. So over time, the long tail just becomes longer and longer and bigger and bigger.

Well, I think the key with 4 in the U.S. is that when you look at the quote-unquote top, a lot of people talk about it like it's a race or something. You were born at the top, but it's not true in the U.S. Almost everything at the top started from next to nothing.

As long as we have equality of opportunity.

So what you should be worried about in bullet 4 is: Can you get to the top? Is there still a way to get to the top? But trying to tax the top and distribute it is sort of not the point. The point is: Do we have an equal opportunity to get there, or have we locked in a certain group of people who control AI and become dominant overlords forever? You want to avoid that, obviously.

All right. So the catchphrase there is equality versus equity.

Five: Why is there no plan from governments for massive job displacement? The answer is there are plans from governments for massive job displacement. Having industrial policies, both in the East and the West, for leaning into robotic automation is very much a plan for massive job displacement.

Six: How do we stop future social unrest? Ibid.

I have an easy answer for that one.

For this one, you don't stop with policy; you stop with agency. Give everybody as much agency as possible, and people feel empowered.

Yeah. The other part is you have to deal with the fear people have. Social unrest comes from fear: fear of not understanding the future, fear of not having a job, and fear of not having a roof over your head. So one of the things we've talked about is: Can we deliver universal basic services to people that give them stability as one of the solutions?

The second thing is addressing their basic concerns. Can I feed my family? Do I have to worry about society imploding on me? People are living in fear. Their amygdalas are lit up by the speed of change right now. We have to address that.

I spent a lot of time in the statehouse, and I can tell you there is no plan on bullet 5.

Yeah, there is no plan. Agree.

I'll take the other position. I think China certainly has a plan, given that China is facing rapid demographic decline while also ramping up its robotics initiatives. The plan is to lean into it rather than be a victim to it.

I will also add that I prepared a movie. I think we talked about it in the past. You can view it on my X profile. It’s called A Nation That Learned to Sprint, where I argue that the way social unrest could be handled in the future is to treat social cohesion as a form of infrastructure that AI itself can help to mold.

But some people find that framing overly authoritarian. So it’s easier to reference other points very quickly. And then, Nate, because these are such wonderful questions: 7. If everyone becomes an entrepreneur, who would we sell to and why?

I think, again, if you analogize entrepreneurship to social media, that’s like asking, if everyone can publish their own essays, for example, who’s going to read them? Yeah, who’s going to read it? The answer is people who want to. Of course, there’s demand for peer-to-peer publication.

8. Wouldn’t UBI destroy people’s personal motivation to achieve in society? I think UBS—universal basic services—is probably more promising, but the point in UBI is the B. It’s basic. So there will always be things and inequalities to hand-wring over and to strive for. I have 2 points I want to make on this one.

One is that the B is really important. We’ve seen that UBI experiments have succeeded when you give people enough to survive but not be happy. So you still have a thriving economy, et cetera.

Yeah. They still want to go higher.

Yes, 100%. But the second point is very important: people confuse UBI with a socialist scheme. It is not. If you implement UBI properly, you dismantle government services—a libertarian scheme—and then you have market forces driving most of it. So this is a really, really important misnomer that a lot of people get wrong. Go ahead.

9. How can individuals compete with large players that can afford $1,000 a day on APIs, which I assume is a coded reference to Dave? It may or may not be the case, but I think it sounds like one.

It definitely is. It definitely is.

And so I would say, follow what I’ll call the China pattern, which is: when you’re compute-limited, be creative, be more resource-efficient, and develop leapfrog approaches that make better use of the compute resources that you have. Also, work on problems that are higher leverage given limited compute.

10. And finally, will our patent and trademark offices collapse under the volume of AI-generated entrepreneurship? No.

I’m going to argue that one, yes. The fact of the matter is, they will.

When we have AI, a patent and a trademark will be meaningless, or at least a patent will be meaningless. AI is what we invent around it.

Remember, IP systems are designed for scarcity of invention, right? AI flips that to abundance, so the whole thing dissolves.

Yeah.

Yeah. I think the rate of patentable things will way outstrip anything the patent office has planned. I don’t know if you could bundle them into groups or something.

But if you’re depending upon a patent as your moat to protect you—

Toast. You’re toast.

Totally right.

For the record, I disagree with that point. I would also point out that the U.S. Patent and Trademark Office also has AI, so it’s not like there’s some fundamental asymmetry here.

Okay.

If patents and trademarks have less value in the future because of the speed of innovation, I mean, you’re either reinventing yourself constantly or you’re dead. We’re seeing that in the whole AI world today, right? Things that were built a moment ago, like SaaS platforms, are becoming irrelevant as Claude 4.5 is reinventing them.

I think it’s a little bit analogous to saying life can’t possibly exist at the bottom of the ocean because the pressure is too great. But what actually happens is the pressure inside the organism matches the pressure on the outside. Similarly, we forget that patent litigators are going to have AIs as well, and the patent office will have AI. Everyone’s going to have AI.

You’re just loving Accelerando so much that—

It is the best book ever.

And Accelerando, if you haven’t read it yet, has a whole segment about the lead character constantly filing patents. Before we leave that slide, on the second-to-last bullet, I had a great meeting with Eric Schmidt, Eric Beinhocker, and Daniela Rus out in Davos, plotting out exactly how we can unleash entrepreneurs and give them access to the compute they need because it is a very valid point.

If you can’t get the best AI in large volumes, it is very, very hard to compete, and it’s only going to get harder. But we have a plan, and we’re super excited. We’ll roll it out very quickly. Eric Schmidt moves fast when he has a great idea.

I love it. I have 2 quick points to summarize this whole discussion.

Okay.

One was the unbelievable conversation at Davos between what’s happening with AI and the geopolitical nation-state conversation, and just the discrepancy and gap between those 2 conversations. The second, for me, is the Claude constitution stuff. It’s some of the most fascinating stuff we’ve seen, maybe in 1,000 years, for the future of humanity.

Yeah.

And important.

Yeah. The same thing. Huge. Amazing.

I, by the way, Alex, you did an incredible job speedrunning these 10 questions. Thank you. Whether you were right or wrong, or we agree with you or not, I love the fact that you took it on.

Lightning round.

Before we get to that, can I just say thank you to you guys? I feel topped up again intellectually.

Yay.

So fun.

Are we going to get to film back in the manufacturing area with Brett Adcock?

I hope so. I’ll ask. It’s so cool, but normally they’re really cautious with that stuff. It’s so amazing when you see them work.

You know my question for Brett.

I know: Why in the world does he have more than 2 arms?

No, no, that wasn’t it.

The opposite. I’m kidding.

All right, guys.

Welcome back from Davos.

I feel the same—recharged and re-energized, and hopeful. I’m coming away hopeful from this conversation.

Yeah. Take care, guys. Be well. Bye.

Thanks, Peter.