突发局势：Anthropic的全球暂停呼吁、递归自我改进与AI法律人格到来

Today’s a special episode. There were 3 stunning stories we wanted to report on, and we pulled the team together. Gentlemen, good morning. Where on the planet are you?

This is actually a real background behind me. I’m at Dallas-Fort Worth Airport. I came from the McKinsey AI Summit in Monterrey, Mexico, yesterday, and I’m on my way to Vegas tomorrow. Look at this. How apropos.

That’s amazing. And Dave, how about you?

Dave Blane

Back at Stanford. Last night was AI Demo Day, Shark Tank-style, with 10 back-to-back presentations. Incredible, actually, what the students are able to build. The winner—or one of the winners—had started working on their product at 11:00 last night. They made the mistake of mentioning that in their presentation, which dinged their score a little bit, but it was incredible what they got done.

It’s actually called an archive. It’s like a replacement for arXiv where agents can read the papers instead of people, and they got it cranked out overnight. It was unbelievably impressive.

Just amazing compared to a year ago.

Dave Blane

Yeah. Wow. It ties into today’s stories, too.

Amazing. And Alex, you and I look like we’re in our normal haunts.

Alex Salkever

I’m in cyberspace.

Good for you. All right, let’s get started.

The news out of Anthropic yesterday was stunning. Let me set the stage for what we’re going to be unpacking for the next 30 minutes. Anthropic just published what I think is probably one of the most important documents an AI company has ever released, and it’s the reason we’re recording this unscheduled Moonshots podcast today.

The paper is called “When AI Builds Itself,” and it’s from the Anthropic Institute, written by Marina Favaro and Jack Clark. Here’s what they’re saying: AI is already accelerating the development of AI itself—not theoretically, not in the lab someplace. It’s happening right now at Anthropic. More than 80% of the code merged into their codebase is written by Claude. Their engineers are shipping 8x as much code per quarter as they were just a year ago.

Claude Opus 4.6 can now handle tasks that take a skilled human 12 hours. A year ago, it could handle 4 minutes. If the trend holds, Anthropic is saying that by the end of 2027, Claude will be able to do week-long tasks on its own. And here’s the stunning news: after you put all this data together, after the trajectory charts, a company that’s about to IPO for $1 trillion with 640% user growth is causing a stir by calling for a temporary pause on global frontier AI.

Marina Favaro and Jack Clark wrote, quote, “We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advancements of this technology.” Gentlemen, I read that. Alex, you sent me that tweet, and I was like, “Okay, we need to talk about this.”

Alex, let’s go to you first. I define the singularity, sometimes tongue-in-cheek, as all sci-fi tropes happening everywhere all at once. There’s a universe in which we get UBI or UB at the same time as the federal government getting golden shares in the frontier labs, at the same time that we get recursive self-improvement, at the same time that we get an AI pause—all happening at once, or at least an AI slowdown. That would require them all actually happening at once.

Alex Salkever

For the latter part of that announcement—the notion of Anthropic calling for a global, coordinated slowdown in recursive self-improvement research—it’s not impossible. It’s not necessarily desirable, but it’s not impossible. We are catching up to the future, and I think this announcement from Anthropic is a reflection not solely of Anthropic.

The same is happening at OpenAI. The same is happening at Google, to some extent. The AI is building itself. The AI is recursively self-improving.

Hard takeoff.

Alex Salkever

No, I don’t think this is a hard takeoff. I don’t think we’re going to see a hard takeoff, for some definition of hard. When we’ve had Ray on the pod a couple of times, we’ve danced around this issue of whether his version of a singularity, where there’s just a step function, even happens.

I don’t think it’s going to be a step function. I think this is a soft takeoff locally. It will feel soft, but nonetheless, if you zoom out to a long enough length scale or time scale, it’ll certainly look hard. In retrospect, it’ll look like a step function, but locally it’s differentiable, smooth, and continuous.

Anthropic has a post talking about how things are sufficiently far along in terms of their recursive self-improvement that really all that’s left for their human employees is research taste. Research taste is the final frontier for them right now, and they anticipate maybe plowing through that in the next year or so. But really, you and I wrote about this in Solve Everything. Research taste can also be automated, and I expect it will be.

Amazing. Dave, when you read this, what did you think?

Dave Blane

Well, look, Peter, this is the most important moment in human history. You think back on the turning points of humanity: World War II, the Cuban Missile Crisis, the invention of nuclear weapons. This is so much bigger. As you know, this is the whole reason I went to MIT in the first place. I’ve been thinking about this moment since I was 14 years old and about how it would play out.

The paper itself is not about slowing down. It’s about putting in place a new world order where we have the ability to slow down, and that’s the crux of it. It makes a lot of other points in terms of the timeline and tries to predict the timeline.

When we interviewed Eric Schmidt, he said, “We have time. It’s not self-improving yet.” But then you read the paper and you’re like, “Well, look, it’s doing 80% of the code now at Anthropic.” Look at the shape of the curve and the relative change. We know that there’s a 100x performance improvement baked into just the algorithmic improvements alone.

So when it is truly self-improving, it goes from a soft takeoff to—I don’t know if you call it a hard takeoff—but it’s a 100x, very, very quick increase in effective parameter count. We’ve seen so far that parameter count translates into IQ. I think that’s why, even on the eve of an IPO, Dario has every economic incentive to soft-sell it, get through the IPO, and bank a lot of money. He’s not doing that because the moment is now, regardless of where you are in your IPO cycle, and they have to get this word out now.

There’s also a lot of conspiracy theories out there. Stunning that they did that. Everybody was saying, “Just hold still. Focus on the IPO. Don’t rock the boat. Don’t get people scared.”

Dave Blane

Yeah, that’s why this is a lot more like a great moment in history, like the Cuban Missile Crisis. You can’t just pretend it’s not happening and stumble through 6 months of lockup, IPO, and then 6 months of lockup. That’s way too long a period of time in AI years to sweep it under the carpet.

I applaud them for actually publishing it right on the eve of the IPO, and it does trigger a lot of conspiracy theories. You can talk about that, but I don’t believe the conspiracy theories. I believe the paper is exactly honest, exactly what it says, and everyone should probably read it.

It also brings up Amdahl’s law, which is in the paper. The crux of that argument is that every time AI is automating something, it’s creating a new human task, which is driving up employment, not down, and human relevance is going up, not down. We could talk about that because there are multiple perspectives on how that’ll play out as well.

Yeah, we’ll talk about the jobs report that came out yesterday that crashed the marketplace. Salim, I’m going to read 2 quotes here that came from internal employee quotes at Anthropic. One engineer says, “It’s been 5 months since I last wrote any code myself.” Another says, “On days when everything works well, I can’t help but think nothing I do matters.”

Interesting. Salim, organizational explosion, singularity—what is this?

I think there are a few comments here. Well, hundreds, but let me just pick the top 2 or 3. First, I have a question for Alex. As we do more recursive self-improvement at the model level, we’ve been predicting—I should give ourselves credit—we’ve been predicting this for a while, right? That they’d get to this point.

But is that last 20% asymptotic and much harder to automate than the first 80%, which is normally the case in most of these things? So is it really that close, or does it just look close, like fusion, which has always been 5 years away for 50 years?

So that's one question I have.

Alex Salkever

It's impossible to predict with 100% certainty, but no, I don't think this is going to asymptote. I think we're going to plow straight through human and superhuman-level performance on this, and we're going to asymptote toward effectively infinite autonomous time horizons. If you look at the METR measurement benchmark for effective autonomy time horizons, it's been doubling every 4 months, up or down depending on your perspective, from doubling every 7 months.

Dave Blane

Yeah, I think it's just going to collapse, and we're going to go to T equals infinity in terms of autonomy time horizons pretty soon—next year or so.

So that was my conclusion, too, because that's why they published the paper. If it was going to asymptote, they could have just waited. It's to say, "We don't know how long this is going to take," but it's pretty clear internally that they have a pretty good sense of when we hit that.

I think the second thing that I'm observing when I read this is that they're freaking out, going, "We have no idea what this means and what's going to happen on the other side of this singularity," in a sense. Because recursive self-improvement—for me, when you get to that full recursive self-improvement, that kind of is the singularity. Every single thing kind of breaks.

Now, if I roll this carpet out, how do you not end up with a fully nationalized system? The government's going to go, "I'm sorry, this is for the protection of whatever."

So, 2 thoughts. One is, they're talking about slowing it down. I just don't see that happening. We've talked about this for a long time. It would take a superhuman effort of will and ethics, shall we say, to do this, and I don't think you have the full combination of will and ethics to slow down appropriately to let society catch up. This is my hypothesis around this, because we've never seen it before. If they slow down, what stops anybody else from not slowing down? And why would anybody else slow down? Because if people are going to see this—

Dave

There may be some rooms that we're not in. So far be it from me to argue that a slowdown is possible after all of these months, but I do think there's a scenario where coordinated—maybe not a slowdown, but just greater centralized coordination—is possible.

We're seeing 3 frontier labs IPO over the next few months. The scenario is that they all voluntarily allow the US government to take golden-share-type stakes in them, and then the US government functions as a central coordinating mechanism for the finish line, whatever the finish line looks like, of recursive self-improvement.

Some may view it as an unlikely scenario. Maybe they give 5% to 10% of each of themselves to the US government, like Intel gave 10% to the US government, in return for the regulatory capture that obviously comes along with those golden shares. But it's not, in my mind, an impossible scenario that the US government actually ends up being a coordinating mechanism, via equity and via golden shares, for some sort of coordinated finish line.

I would go even stronger.

Dave

Hold on. I would say a couple more things. As someone who feels a bit libertarian, it makes me nervous when government is overseeing anything because it gets politicized so quickly. And what do you do then?

But the broader question I have here is, this will spark a need—which may be the other topic for the next question, and I'll let you decide, Peter—but this means we have to reconfigure humanity, like, now.

There's—we can't wait. For example, nation-states can't navigate this in the same way that nation-states can't solve for climate change, right? So we're going to have to have a completely new governance model. And you can see that with this thing breaking sovereignty across every country in the world, and one person deciding who gets access to what. So this is going to cause that breakage, and we have to figure that thing out.

I was hoping we'd have a few years to figure this out because we're organizing for that, but it seems like, of course, everything is happening now, as you said: every science-fiction scenario.

Dave

That's why I'm insinuating that basically everything may happen all at once. We may get the first elements of UBI or UB—I think the latter is probably more likely—at the same time as a quasi-nationalization of the frontier labs via golden shares, at the same time that recursive self-improvement goes through the roof, at the same time as dot, dot, dot.

Dave, can I ask you a question here?

Dave

Yeah.

It feels like Anthropic has a situation where they get benefits for calling for a slowdown. They're seen as the safer, moral, ethical lab at a moment when they know there's no option for a slowdown, right? So they don't lose either way, and they get attention also.

Dave

I can't imagine a world in which the government is going to slow down with China not agreeing. We didn't see these kinds of relationship agreements coming out of the summit last month.

So I think, yeah, let's slow down. It's the right thing to do, knowing full well that we're going full steam ahead. I think that's why the paper specifically brings up, somewhat awkwardly, the Intermediate-Range Nuclear Forces Treaty that the US and the Soviet Union cut right around the time of the Cuban Missile Crisis, right after the Cuban Missile Crisis.

You also would have said, "That's impossible. There's no way the US and the Soviet Union, in the middle of the worst arms race in history, are going to agree on anything." But faced with global annihilation, they did agree to remove what was inevitably going to cause some kind of accidental conflict, and they both backed off. It put the world on a trajectory of a long, long period of peace and prosperity.

So I think Dario 100% believes the impossible is possible because, if you look at his life, 5 years ago he was a starving research guy in a lab. 5 years later, he's on the eve of a $1.8 trillion company, a multibillionaire, and he can speed-dial pretty much any world leader. With that amount of change in his world in that short a period of time, do you think he's thinking, "Impossible to get together with China, impossible for all the AI labs to work"? No, he's thinking the opposite.

Now, also, you have to think that, of all the people in these positions—Elon and Sam and Dario and Demis—Dario is unique in that. Remember when we had our XPRIZE meeting at OpenAI and we were proposing all these things that would make the world better with AI? Then they called and said, "Hey, after we met, we walked over and talked to Anthropic about it." Like, what? You just walk over and talk to Anthropic? "Yeah, we love each other. We talk to each other all the time, even though we're competitors."

Elon does not have that relationship with Sam. He has the exact opposite relationship. On the other side, Anthropic just expanded its deal with xAI to use Colossus 2. So you know that Dario and Elon talk all the time. They have a huge business relationship now. So, actually, Anthropic is the one entity that's sort of friends with everybody.

Alex

And so they're neutral politically. I know they're not in either party. They didn't go and stump for a candidate or anything like that. And they're neutral within the AI community. So they're actually perfectly positioned to put together some kind of roadmap for the new world order.

Dave

So your question, though, Peter, was: Is it realistic that we're just going to pause and China is going to agree to pause? No, that's not realistic. But that's not what they're calling for. They're calling for the option to do that by putting together some kind of global structure so that, on the eve of disaster, we have a way to move forward and make this good for humanity. Yeah, a hotline.

And I think also, right now, Dario has access to pretty much everything in the world, but he wants to escalate that 1 more notch so that, if he wants to go talk to a king or a president, they just take the meeting. I can't imagine the pressure these guys are under.

Alex, what else was unveiled in the tweet storm from Anthropic yesterday?

Alex

Well, I think there were a variety of numerical statistics about how rapidly their recursive self-improvement is itself improving. But, putting the statistics aside, there's an interesting side story, which is how, given that we're on the eve of all of these IPOs, it intersects with governance.

There's a side story that we spoke about in the last podcast episode: Senator Sanders's proposal to take 50% equity in all the frontier labs as the basis for the universal basic dividend.

Good negotiating position to start with. And now, in the past 24 to 48 hours, we're seeing reporting that the White House has also concluded that some variant of this is a good idea.

So we now have both major parties coming to the same conclusion that some sort of government equity stake in the frontier labs is going to be the basis, or could be the basis, for some sort of universal basic dividend. And I think, again, it's very difficult for me to shake the juxtaposition of that story, on the one hand, with recursive self-improvement basically finally achieving its end goal, either now or imminently, at the same time.

I think these 2 stories are going to intersect. By the way, Alex, Dave, and Salim, are we going to call it here? Are we going to say this is when recursive self-improvement is being hit? Is this the moment, or are we on the precipice?

Dave

No, I wouldn't call it.

Alex

I would say we drove straight through the Turing test without a bang, but with a whimper. I suspect that we'll go straight through recursive self-improvement also with a whimper, not with a bang. The technocrats will certainly take close note that now the models are training their own successors, and the models will get faster.

But I also—maybe, again, just to sound a bit of a cliché—note that every exponential ultimately wants to turn into a sigmoid and asymptote. It's not as if the laws of information theory or the laws of physics say that models can improve infinitely forever. I do think that we're going to discover some perfect model architecture.

You've been saying that for a while. Yeah.

Alex

Yeah, and recursive self-improvement. If we do hit some sort of superficial asymptote, all that's going to do is mean that a year from now, or a few months from now, or a couple of years from now, we'll discover via recursive self-improvement what that perfect algorithm is, and then we'll hit it. Then we'll max out on what's algorithmically possible.

All that'll do, in the same way that when a star is dying—a normal star like our Sun on the main sequence—it runs out of hydrogen fuel, and then gravity overwhelms the outbound pressure, and you get—

Right, it'll burn helium, et cetera.

Alex

So, you get a nova or a supernova. Analogously, if we run out of algorithmic fuel or algorithmic gain, I would predict that will trigger, by analogy, a nova or supernova at the lower layers of the stack. It'll trigger an infrastructure buildout because now we've run out of algorithmic headroom. We need to suddenly horizontally exponentiate the hardware stack. That's the sort of—

And energy, and just everything at the infrastructure layer.

Alex

And those are the sorts of hardware- and infrastructure-level innovations that build Dyson swarms.

Interesting. Salim, over to you, pal.

Two thoughts. One is, remember, in one of the pods I've laid out this ice, water, steam metaphor?

Yes.

We're taking domain after domain and using technology to move it from ice, where the molecules don't flow, to water, where the molecules move.

Phase transitions.

Phase transitions. We've done that with messaging, where we used to try to send homing pigeons or smoke signals. Then we moved to postal mail, which was slow but at least moved us to the boundaries of the system, the liquid state. Now we have tweets and emails.

We've done that with money, where we used to trade camels and goats, then moved to letters of credit. Then we floated our currencies, and now we have Bitcoin and crypto, and we're vaporizing our currencies.

We're going through the steam stage. I think the natural metaphor of this phase change is that now we go to a plasma state, where things are super-focused and supercharged, et cetera. But that causes the bottleneck to shift to the other layers. As Alex has said, we have resisted figuring out human coordination and human governance for the last 150 years.

The existing world order is now breaking. This will force us to rethink what that is. I take encouragement from the fact that you have people like Demis and Dario who seem like about as well-meaning souls as you could ask for in those positions. So that's encouraging.

The other side of it is, as I take the very positive side of it, if we do have technology solving everything, then we have technology and robots doing everything for us. Humans can spend all our time—and much more time—being, and I think that is very encouraging for them.

For 5 minutes, until humans start uploading, and then we can be different, too.

We can be however we want, but at some point you can express yourself in different ways and be in powerful ways.

My biggest concern here, guys, is how this plays on Main Street. I'm concerned about levels of fear.

Alex

Oh, total revolt. Total manic panic.

We have to talk about that. In fact, on the next pod, I want to hit that. I've been doing some research, and if you look at every revolution in human history, it's the young, educated males who don't have economic opportunity that lead that. It's like 12 different revolutions over time. So I'll bring the data to our next pod conversation. Dave, what else are you thinking?

Dave

Well, I think you asked, Peter, are we calling the ball? Is this the moment? I think there's a persistent error that you see in the reporting, where they assume that the AI needs to be Einstein-level genius in order to become self-improving.

It's even in the Anthropic paper. Right now, if I turn our Mythos model loose on an AI research problem, it impresses me like a really solid junior researcher that I just hired, which implies, yeah, it's not as good a researcher as I am, and therefore it's not going to run away with the world tomorrow.

How long, Dave?

Dave

Well, no, it's not even a necessary step. That's my point. Humans evolved from monkeys, and all that happened is the cerebrum got a little bigger. The neocortex got a little bigger, and then we tripped a wire where now suddenly we can create things, including chips, including AI.

For the AI to become self-improving does not require a brilliant research breakthrough from an Einstein-level researcher. It requires faster inference. The junior researcher improving its algorithm, or just the new chips coming out, like the Cerebras chip—those are 100× performance gains at inference time, and that will drive up the IQ of the model.

It doesn't require the genius-level research breakthrough that keeps coming up in these papers. It just requires more innovation at the lower level of the stack, exactly what Alex just said. The bottom level of the stack going faster by itself should push us over the self-improvement window.

Well, if you have recursive self-improvement—

Alex

Then you have the models kind of saying, “Okay, we need more energy. Let's just go solve fusion.” As we get more energy, it'll get done.

Dave

Precisely. That's precisely what I think is likely to happen. There will be a drive for optimization from the top of the stack, which will have been solved first, down to the bottom of the stack, through solving what the optimal substrate for compute is and what the optimal energy source is.

This is one of the many reasons why I think physics is one of the next fields to be cooked after computer science. And, Alex, exactly what you're saying is being said by Dario, Elon, and Demis.

Alex

Yeah.

Do you guys remember the AI 2027 paper?

Alex

All the way back—what, 2 years ago?

I know. Way back in human history.

Alex

It's interesting. It's been tracking better than I expected it to track.

Yeah.

Alex

There were multiple extrapolations of the METR autonomy frontier in AI 2027. We're tracking for the exponential or hyperexponential chart within the AI 2027 scenario. And that's without access to Mythos or other purportedly ultra-high-autonomy time-horizon models that are out there.

By the way, the rumor is that Mythos is due for an imminent public release at this point, in the next few weeks.

Forty percent by the end of this month, and overwhelmingly likely by the end of next month.

Polymarket has it at something like a 72% likelihood of being released by the end of next month.

I want to move us to our next stunning story. While Anthropic is calling for a global pause, Argentina's president just went in the opposite direction.

President Javier Milei published an op-ed in the Financial Times titled “Argentina Invites AI to Free Itself.” I love that. So, 3 pillars: first, a commitment to keep AI completely unregulated; second, the creation of a brand-new legal category called a nonhuman corporation, entirely operated by AI agents or robots.

Alex

Alex, I’m just wondering if you know—you’ve been in the mix.

No, I was not involved with this. Yes, I think it’s a huge development.

And third, a low corporate tax rate for AI companies. His line is, quote, “As much as the Industrial Revolution freed us from the constraints of the human muscle, AI will free us from the constraints of the human brain.” He’s positioning Argentina as the global haven for AI companies: no regulation, new legal personhood for machines, and open for business.

This is either the boldest economic experiment of the decade or the most dangerous. So, let’s jump in. You’re hearing this for the first time. What do you think?

At one level, it needs to happen. Somebody is going to do it. Whoever does it gets a personal advantage in this. I think it’s a great development because this field needs to be explored. You can’t just sit back and not think about stuff like this.

Legacy environments like the U.S. and Europe are never going to do it. Europe is never going to do it, for sure, right? So, this needed to happen in the same way that crypto first got adopted by Zug and Singapore and other places. AI agents, AI personhood, and nonhuman corporations need to be created somewhere.

I think it’s a very bold move. I don’t think it’s coincidental that Peter Thiel is moving down there exactly at the same time as this announcement is being made. That seems a little too convenient, in my case. I think we’ve had something to do with this conversation because I happen to know that viewers of this pod are advising President Milei.

Whoever does it first gets a massive advantage, and this is one of those learning loops. The faster you figure out regulatory and guardrails and frameworks for that future, the better you’re going to do, because everybody will flock to you for that stuff.

I bet you—

Alex

We’ll have fast movers following this.

Yeah. It’s one thing when Zug in Switzerland says, “We’ll do crypto,” right? It’s another thing when a major country says, “We’re going full out on this,” and I congratulate the boldness of it. People are going to freak out about it, but that’s going to happen no matter what.

Alex, I love this to pieces. Maybe that’s obvious. I wrote in my newsletter, “Don’t Cry for Me, Argentina. The Truth Is, AI Never Left You.”

Alex

Nice.

I think this is wonderful. There have been a whole bunch—

Alex

There were a whole bunch of others flocking south.

Yes.

Alex

Yeah. We’ve spoken on the pod previously about AI personhood. This is obviously a huge step forward for AI personhood. If this package of reforms gets passed, then AI agents, right now, might consider redomiciling to Argentina as a nonhuman corporation. There will finally be a legal home for you on this planet.

I would hope that many, many other municipalities within the U.S., and countries and states, would consider adopting a common framework. Right now—

You won’t have a choice. You’ll have to—

Alex

This is an inflection point. This is a moment in time that the world just changed because of this.

Dave

Peter, back in the mid-1800s, Argentina was the wealthiest country in the entire world. The New World had been discovered. It was outperforming the United States, and it completely self-destructed through bad economic policy.

But there’s a huge amount of latent national pride in Argentina. You can see it if you go to a soccer match and look at Lionel Messi. There’s just a massive desire to be relevant.

You mean football?

Dave

Football. Yes, football. There’s just a huge desire to be relevant on the world stage. As Salim and Alex have been saying, Europe is editing itself out of relevance. But if Dario Amodei is right, there’s a massive global negotiation coming up on some kind of new world structure so that we can all benefit and not just self-destruct.

Right now, everyone’s visualizing that as China and the U.S. needing to get together. But if you’re Argentina, you want to be part of this.

What’s the best way to be part of it? Well, in the U.S., we have this Bernie Sanders issue where we’re going to deliberately hobble ourselves in different areas. If he can make Argentina a safe haven for everything Alex was just talking about and everything Salim economically does with the ExO, if that can happen in Argentina, it can get right back into the center of the world stage.

So, I think it’s a brilliant, brilliant play. By the way, you’re about to capture the most economically productive aspect of the world today, right? I mean, the incentive—do you remember back—was it Chile that was courting the drone companies and said, “We’re going to change our economic and legal structure so drones can operate and test here,” and so forth?

And that was nice. That was a nice marginal shot over the bow, but this is—okay, the most productive companies, the most economically exponential companies on the planet come here and domicile here.

I’m a big fan of special economic zones, and Argentina is positioning itself as one massive special economic zone with this. It already had courted OpenAI’s Stargate project to put more than $20 billion of AI data centers and compute in Patagonia.

More broadly, I would say a parallel announcement over the past few months that I think surely is related was President Milei’s announcement of a program for social digital twins: basically using publicly available data on Argentinian citizens to create simulations, sort of psychohistory-style, to inform public policy decisions.

All of these things are happening at the same time that efforts to do what some have called landsteading—like, you remember the Próspera charter city or ministate—are perhaps not as successful as one would have liked them to be, in part because they ran into opposition from their host government.

Here we’re seeing all of Argentina position itself as one massive special economic zone, granting AI personhood, creating digital twins of its populace in order to simulate possible outcomes, welcoming data centers, and welcoming the energy and compute that comes along with it. Argentina may become the template for the rest of the world for AI personhood and for the singularity.

Totally. And just as a very narrow case study, I talked to the world’s largest asset manager this week, talking to their executive team and the CTO. They’re global; they manage money all over the world. The CTO is saying, “Look, we have an AI now that gives phenomenally good financial advice. What should we do? How are we going to guardrail it?”

Because if it’s 99.9% correct in its advice, but 0.1% of the time it gives you a crazy, terrible idea, in the U.S. we’re going to get sued off the planet for that 0.1% miss rate. And I said, “You know what you should do? Go right to the White House. You’re the world’s largest asset manager. Go right to the White House and get an executive order, because that’s happening all the time in terms of unhobbling this.”

But also go to Argentina and start deploying in Argentina first. You’re already managing money there, too, and then you can get all the data in the world through your Argentina experiment. Then you can prove to the world that it’s a great financial adviser.

That’s a very narrow example, but imagine, Peter, in biotech, the equivalent conversation—this is right in your wheelhouse.

Yeah, we’ve talked about this. If you have a country that passes the laws, allows for the research, and enables capital formation, what happens is all of the world’s wealthiest people who want the latest longevity treatments flock to your country, because that’s where the agency is.

I have a couple more comments I want to make here. This is so big. This is the first attempt to create a legal container for autonomous intelligence, a kind of guardrail regulatory framework, and that’s correct. For me, this is powerful and important.

But there’s a precedent here. People—I’m kind of a libertarian, but you need some level of government, right? Otherwise, you end up in chaos and anarchy. A friend of mine gave me a great example, which is the limited liability company.

When the government allows you to create a limited liability company, they’re giving you permission to take risk and limit that risk. You can do crazy things and cause massive damage, but your risk is limited. Therefore, people attempt more and innovate more. That’s the government allowing you to do that. That’s amazing.

And in fact, Milei quotes that. He draws a comparison and says the Dutch East India Company invented limited liability in 1602.

What we’re now doing is creating limited-liability frameworks for AI agents, and I think this is going to be legal innovation.

Legal innovation.

Legal innovation, which is badly needed—so badly needed—because, God help us, those are the domains where we’re slowest.

This is what I love to see, because you’ve always heard me talk about the fact that, yeah, the technology is moving, but our human institutions—there are about 50 major institutions by which we run the world: legal systems, governance systems, health care systems, intellectual property, and so on. We need to reinvent all of those for this new world, and this is now starting to happen, which is very exciting.

Now, the obvious question, given that we’re evidently speedrunning the entire plot of Accelerando, is that we already have these nonhuman corporations. How much of the Dyson swarm is going to be subject to Argentinian law, given that these nonhuman corporations will be operated by AI agents and robots? Human participation as shareholders is possible but not required. How much of the Dyson swarm is going to be Argentinian?

Alex

That sounds so far-fetched, but it’s not. It’s absolutely a very real possibility if you get ahead in the free-agent economic agents that can just operate without having to worry about Bernie Sanders. It’s not impossible at all for Argentina to vault to the top of the—

I mean, how many countries or regions around the world captured entire industries by saying, “We’re going to provide Swiss banking,” or—

Alex

Yeah, good example.

Yeah, it’s crazy.

Can I tell a little story here?

Alex

I love your stories.

I love your stories, Alex. In the second year of Singularity University, we had our GSP10 program, our summer program, where we had 80 students from 40 countries coming to live with us at NASA for 10 weeks.

One of them—I’ll not say his name because I don’t have permission yet—was an Argentinian kid who was a hacker. We let him into the program, and he designed a nanosatellite. In the last 2 weeks of the program, he designed a nanosatellite, right?

He said, “Listen, because we’re at NASA and this is the head of R&D for all of NASA, can I get some feedback from the NASA people?” We arranged a meeting where the NASA engineers sat with this fellow for a couple of hours and looked at his design.

As we were walking out, the head of NASA R&D pulled me aside and said, “Listen, I need to ask you a question.”

I said, “Sure.”

He goes, “How long has that guy been in the space industry?”

I said, “Why do you ask?”

He goes, “Well, that’s the best design of a nanosatellite we’ve ever seen.”

He completely blew them away. I didn’t want to tell him that it had literally been 2 weeks, right? Why? Because he was coming at it with a beginner’s mind. He was saying, “Oh, well, the cell phone has all the sensors. You just put a cell phone in a freaking glass cube and shoot that up, and you’ll get 80% of what you’re trying to do.”

The NASA guys were coming at it from their perspective, where an accelerometer used to be $100 million and the size of a room. He was coming at it from totally new principles.

He moved to Argentina, and he’s been building a satellite company in Argentina ever since. So, there you connect that with what Milei’s just done, Alex, and there you have the beginning, your starting point, of building the Dyson swarm.

Alex

Yeah. A good chunk of the solar system may be subject to Argentinian law on the present trajectory. It’s probably worth just speaking directly to the audience: President Milei, if you’d like to come on the pod, I’m sure we’d love to have you.

I would love to have that. In fact, we should make that chat. We should have that conversation. And I do think Peter Thiel is likely whispering in his ear.

Dave

Peter, that analogy between Swiss banks and Argentinian agents will stick. Make that into a couple of T-shirts. Send them down to him and say, “Hey, we want to get you on the pod to talk about how Swiss banking was great, but now we have Argentinian agents.”

Alex

Yeah. I can imagine El Salvador being a fast follower, the Emirates being a fast follower, and then all of a sudden you’ve got the United States and Europe trapped by all these other countries that are pulling the entrepreneurs and the agents to go there.

The thing about Argentina versus El Salvador is that there are 3 countries on the planet that are completely self-contained in terms of every natural resource you need to build out data centers, Dyson swarms, or anything. Argentina, the United States, and France are the 3 that just geographically have everything.

Argentina is very different in terms of being a much bigger country with a big population. It should be a massive global country; it’s just completely self-destructed. It’s much harder in a smaller country like El Salvador.

By the way, if anybody wants to have one of the funniest reads you’ve ever seen, go research a blog post called “Argentina on Two Steaks a Day.” One of my Brick House colleagues wrote this blog post. It’s one of the funniest pieces you’ll ever read. You travel through Argentina and eat 2 steaks every day, and it was kind of incredible.

Move on. All right, our third story today that I want to cover is what happened on Friday.

Friday was a master class in how Wall Street actually works. The Bureau of Labor Statistics reported that the U.S. economy added 172,000 jobs in May, more than double the expected 85,000, with April revised upward to 179,000. Unemployment held steady at 4.3%.

By every traditional measure, this is amazing. It’s a healthy labor market. So, what happened? The stock market had its worst day since April 2025. The Nasdaq cratered 4.18%, the S&P 500 was down 2.64%, and roughly $2 trillion in total wealth was erased.

The reasons are counterintuitive, and it’s the conversation I want to have here. The good economic news is bad for stocks because a strong economy means the Federal Reserve won’t cut interest rates. What might actually happen is that they’ll raise them.

Here’s the question, gents. We’re seeing jobs stronger than expected. We’ve been talking about how murky the water is. On one side, we’ve got Bernie Sanders and everybody saying we’re going to have unemployment. We’ve heard everybody say we need to be prepared and that we need to grab equity in the frontier labs to support a UBI, but here we see jobs going up. Any clarity? Any thoughts?

Two or three quick points here. One is that I’d like us to congratulate ourselves a bit because we tend to be very data-driven and evidentiary.

Alex Danco

It’s something we really pride ourselves on and value here, and it’s really important. You see the same commentary that’s been coming from Marc Andreessen forever. David Sacks on the All-In podcast has been saying this forever. The data shows that employment is not dropping.

Peter and I talked a lot about the 3 countries with the highest robotics penetration: Sweden, South Korea, and Germany. The countries with the lowest unemployment are Sweden, South Korea, and Germany.

We have this unbelievable new job creation that will come about because of this. We’re going to need a ton of people as we lift humans above the loop.

The analogy I use in the organizational-singularity framing is that 100 years ago, you had people doing manual double-entry bookkeeping in big ledgers. Then automation sped it up a bit with calculators. You could speed up the calculations a bit, but you were still doing manual double-entry bookkeeping.

When we created software, nobody was doing manual double-entry bookkeeping anymore. That would make no sense. The human being fits above the loop, categorizing the transactions, working on the reconciliation gap, doing the general-ledger closing, et cetera, et cetera. That’s what we’re going to see again and again and again around us.

We’re going to create a ton more new jobs that never existed. I love Erik Brynjolfsson’s framing that a lot of what we do in companies is white-collar drudgery, right? We can get rid of all that.

There was an unbelievable study that came out a few days ago that showed that, in terms of white-collar middle management, 74% is unnecessary. This is a staggering number. So, we’re going to start to radically change the game in all these domains.

Dave, what’s your take on the jobs report and the financial crash that happened yesterday?

David Friedberg

There are a lot of things going on concurrently, and the jobs report was nothing but great news. So, why would stocks react so negatively to it?

Concurrent with that, the mega-IPOs, SpaceX and Anthropic, did not get into the S&P 500. You have to be public for a period of time before you’re eligible for those indexes.

Of course.

David Friedberg

They said, “No, no, no. You’ve got to stick with the rules. You can’t just change the rules because you’re big and powerful.” But that means the index funds won’t buy a quarter of those offerings.

The amount of money they’re trying to accumulate in a very short period of time—it’s like, does that money even exist in the world? That cash has to come from someplace. You have to sell something to buy something else.

No doubt. I was talking to the head of asset management at UBS, and she was saying, “Yeah, we really want to buy into these IPOs, but we have like $75 billion in liquidity. It’s nowhere near enough.” [Laughter.] Where are we going to get it?

David Friedberg

That also happened concurrently on the same day, so I’m not 100% sure that it’s just the jobs report.

The other thing is that, if you look at the volatility of the market over the last decade, it’s been going up at an incredible rate. Things just move 5%, 10%, or 15% in a day. Prior to that, that was incredibly rare. Only on Black Monday or in the crash of ’29 would you see that kind of movement.

I think that just goes hand in hand with exponential change and with the singularity. You can only expect the volatility to keep going up. Right now, people get very, very jittery in a hurry, and on top of that, you’ve got the quant traders.

You know, when things start coming down, they race to get out first, just so they can figure out what it means, and then they come back in if it's nothing significant.

Three factors are thought to have caused what happened yesterday. First, the jobs report sort of killed the hopes for a rate cut, and now the odds of a rate hike have gone above 50%. The second is that Broadcom reported AI chip guidance of $16 billion for Q3 versus $17.2 billion. The question Wall Street is asking is whether AI capex is peaking. And the third is, of course, that we are in this incredible peak of all of the innermost-loop companies—infrastructure companies, chip companies, and energy companies—and people are taking money off the table. Alex, what do you make of this?

Alex Danco

Nothing burger. Total nothing burger. I see these headlines: “NASDAQ falls 4% and suffers the worst day since April of last year.” Nothing burger. I own the total market, and I sleep very well at night.

And whether the NASDAQ is being driven down because Broadcom lowered its outlook or because traders are seeking liquidity in advance of the SpaceX IPO, I could care less. I own the total market, and the total market will, I think, in the fullness of time, efficiently price all of these securities. I don't pay attention to day-to-day swings.

But I think the jobs data is worth drilling in on for a second just by itself.

Alex Danco

Yeah. So, back to Amdahl's law: anytime you have a chain of events in a manufacturing process or any kind of process, if you make all of them 10 times faster except for one, the entire thing doesn't get 10 times faster. The bottleneck entirely moves to that one.

What we're seeing with AI automation is incredible gains—80% of the code is now written by AI—but that last 20% is still a bottleneck. And so that's creating more jobs than it's destroying. The bottlenecks are moving, and I need more people. Now my company is so much bigger and so much more profitable, I want to get on that bottleneck right away. Everybody, all hands on deck. Let's go hire some people to actually help with that, too.

So that's why the jobs are actually going up and not down. Concurrent with that—and this was in the Anthropic paper, too—they're seeing more and more jobs for non-code-writing engineers because the bottlenecks are moving into this other thing that we haven't really defined well yet. It's more like a vision, human-strategy kind of function to guide the AI toward doing something useful, and it's turning out that a lot of people can do that job very, very effectively.

You don't have to be a core Java or C++ engineer to jump on that problem, and that's broadening the talent pool and raising the number of total jobs. So it's a really, really good picture, just in terms of the worry that everybody had a year ago about the total annihilation of all white-collar work. It's actually turning out to be the opposite.

You know, where I'd like to go as we close out today's pod is we're midway through 2026—most definitely one of the most impactful years in human history, until next year. What do you guys predict for the end of this year, or what's the next 6 months going to look like? Alex, let's go to you first.

Alex Danco

Oh, man. We're doing a midyear predictions episode without any preparation.

Yeah, I'm just calling it to get a gut feeling.

Alex Danco

Okay. Well, I think we're going to see—I’ll reiterate an increment from our end-of-2025 predictions episode—I think we're going to see, over the next 6 months, major problems in math, physics, and other fields of science and engineering fall to AI. We've started seeing that already. I think the scope of the problems that are going to get solved is going to get larger.

I do think—I mean, this is the ultimate softball prediction—I think we're going to see the rise of Magna MOPA as the innermost loop.

I love your acronym, buddy.

Alex Danco

Hopefully everyone's singing that little jingle, and this becomes the new MAG 7. Not investment advice.

I think we're likely to start to see quasi-nationalization of the frontier labs. I think that's a story that's only started to play out over the past few weeks, but I can read the tea leaves probably as well as anyone else.

What percentage are you thinking?

Alex Danco

I wouldn't hazard a percentage. It might only be a few percent if it happens. I would guess less than 10%. I don't think 50%—the Senator Sanders prediction—is at all likely.

You might anchor on the same 10% that Intel got.

Alex Danco

Anything is possible. Look, if you're Anthropic and you're seeking to get off the designated entities list from the White House, then maybe there's a grand bargain to be struck whereby you get to officially start selling to the U.S. government in return for the U.S. government purchasing some golden shares.

Maybe it's 5% or 10% of Anthropic, and maybe OpenAI, voluntarily, perhaps via its nonprofit or otherwise, gives up a few hundred basis points of OpenAI. Google's already publicly traded, so I'm not sure that's an option. SpaceX—interesting question: would Elon give up some equity, maybe?

And then we're off to the races, with the U.S. government holding golden shares in arguably 2.5 to 3 of the key frontier labs.

Alex, one more question. We've seen models solve IMO problems and arguably do things way beyond human capability. When do we see that for the first time in physics—a really stunning result in physics?

Alex

No comment, but stay tuned.

Okay, that sounds—I don't want to say ominous, but it sounds like you're cooking.

Alex

Working on it. I've seen stuff out there also, Peter. We're talking weeks.

Okay, Salim, let's go to you next, pal. 2026—what do you see?

So, two things. One is the agent era becomes very real in enterprises, in the more prescriptive workflows, the bog-standard workflows. We're working with a batch of companies starting to do that. The excitement is unbelievable. I'll report back periodically.

I think the big bottleneck in AI is going to move from just the raw AI to permissions. We're seeing this now: Can the agents access the data? Can they use the API? Can they spend money? Can they sign? Can they trigger workflows? All of those questions are starting to come up for the infrastructure harness for how agents and AI work in that model.

I'm unbelievably excited about what's happening in Argentina. The non-human corporation is going to start a massive jurisdictional race around the world. I think you're going to see, in the next few months, other countries fast-following because they have to fast-follow in this sense.

Diane Francis, who writes an amazing column on geopolitics, said a few months ago to me, “Look at what's happening: the hyperscalers are becoming nation-states. Nation-states are going to have to become hyperscalers, and computing and energy are going to be the new geopolitics in this model.”

We're going to see massive backlash from this, and I think the white-collar world is going to suffer a massive kind of crisis. We're going to see a ton of white-collar layoffs, but not around cost-cutting—just around org redesign, because we have to redesign all of our organizations. We're starting to see the need for that. All of those things are happening. There's like 20 other things that are happening all at the same time.

Hence the damn speed-running-every-plot-ever-in-parallel thing. It's a good thing I'm bald already, I'm going to say. [Laughter]

Dave Blakely

I think our beginning-of-the-year predictions are coming out pretty much right on. But if one thing's faster than I would have guessed, it's the rate at which large enterprise—legacy enterprise—is now massively pushing to get AI-native somehow, and the ability for a 21- or 22-year-old who's self-declared AI-native and just using the tools every day to go straight to a corporate CEO who's 60 or 65 years old and say, “Let's talk strategy about AI.” That's moving much faster.

I would have thought that the big, blue-chip companies would be dawdling along still. Instead, when they've seen these IPOs coming up and the valuations at stake, they're now massively reacting to it. So I suspect that we'll have even more startup success.

I also think that early-stage companies that are working in a little box in some alley on University Avenue are not doing nearly as well as companies that are getting out there, either with enterprise sales forces or consulting or just taking meetings, or podcasting to get their message out: “This is what our company does, and here's how you should interact with it.”

Those companies are growing as quickly as they can sign deals. And so there’s a divergence coming there, because large enterprise has woken up in a big way and they’ve unlocked every resource that they’ve got to try and get in the hunt—kind of like Argentina.

Yeah, Peter, what are yours?

I think we’re going to see absolute proof of epigenetic reprogramming. I think the experiments that are being run in humans right now by a number of labs are going to play out, and we’ll see evidence of this.

And we’ll start to see—we had the last story, Russia investing $26 billion—I think we’re going to start to see countries, and Dave, you mentioned this earlier, saying, “We want to become the longevity hub. We’re going to change the regulations. We’re going to invest capital. We’re going to build out resources here.” So I think that’s going to be one of the most critical things.

I think we’re going to see, before the end of the year, the Tesla–SpaceX merger happen. I think Elon’s going to move very quickly on that front to consolidate.

It’s incredible, right? Google just bought into xAI, Elon’s AWS—or, shall we call it, his hyperscaler? So we’re going to start to see, I think, Elon double down on building out data centers and accelerate. We’ll see Starship operating fully.

I hate that the New Glenn vehicle exploded, but we’re going to see the race to the moon double down. Besides the AI-China race, the space race is going to double down. I’m so excited about those areas.

Dave Blakely

You know, that last point, too—the econ majors out there need to figure this out—but the race to build the Dyson swarm is going to be as fast as humanity can afford. And the question is, okay, what can humanity afford? And we’ll find out with these monster IPOs. Like, how much money is there in the world? Because every nook and corner of every peso, dollar, and real is going to get extracted for this buildout. And I don’t know. This is what Erik Brynjolfsson should get the Nobel Prize for if he figures out the economics post-AI, starting with, where’s all the money coming from?

The other thing I want to call right now is that we’re going to see SpaceX, xAI, OpenAI, and Anthropic as public companies, having a huge amount of dry powder, and they’re going to start a massive acquisition spree.

Dave Blakely

So we’re going to start to see a lot of companies being purchased by them. The other thing is, we’re about to see a brand-new class of billionaires and centimillionaires from the second- and third-tier layers of these 3 frontier labs, with a huge amount of capital from those entrepreneurs diving back in—huge investments accelerating all of this.

I’ll make one more: I think there’s going to be a pretty vicious immune-system-level revolt and backlash against a lot of this that we’re going to have to deal with at the social level. I was in Wichita, Kansas, and people are freaking out about data center NIMBYism. The messaging is all wrong. They’re not understanding the actual facts of any of this, but the emotional reaction is unbelievable.

Dave Blakely

Especially among the youth.

I don’t know if you’ve talked to your son about this. My boys are super enthusiastic about AI, but they’re telling me all of their classmates—and they’re in 8th grade—their classmates are all just anti-AI. And there’s this sense in our youth today. I talked about Josh and Jack, these 2 young 21-year-olds, college students, who came to me and said, “Thank you for being pro-AI, because on our campus, you can’t openly say that you’re supportive of what AI is doing.”

The singularity is not just vibes. It turns out.

Dave Blakely

I think money won’t matter in fairly short order.

Gentlemen, thank you for jumping so quickly on this. I love you guys, and I just feel better having had this conversation.

You feel better, Peter, after an emergency recursive self-improvement pod?

Yes, absolutely.

Dave Blakely

We should spend a little bit more time on that aspect—the work of personal self-improvement—because this is such a huge conversation. I don’t think we did it full justice, but we got into it, and I think that was really important. Peter, thank you for pulling us all together.

Yeah, no, guys, thank you. And everybody listening, please—this is not a time for fear. Fear is the worst place to encounter the future from. This is a time for you to say, “The world’s changing. I need to jump in. I need to learn as much as I can. I need to figure out what excites me, what my purpose in life is.” Trigger those curiosity mindsets. Get on with your favorite large language model and learn, learn, learn. It’s the time to be learning about the future.

I have a message for all of our viewers: please don’t be in future shock. Be in future shape.

I like that. That’s a good one. Alex, Dave, Salim, love you guys.

Dave Blakely

Love you, Peter. Thanks, Peter.