Dario为自己辩护、数据中心恐慌、AI末日主义陷阱、参议院悬念

Dario Amodei has spoken. Apparently, he listens to the All-In Podcast, and he was compelled to respond to Gavin Baker's discussion last week that many people inside—or some people allegedly inside—Anthropic believe “that they will be the only private company in the world at some point.” Sholto Douglas from Anthropic fired back, “Completely false. Whoever told Gavin this is lying.” Dario decided he would come out of hiding and entered the chat with an epic two-part essay.

On regulation, Sacks, he said, “I know the Silicon Valley shorthand where regulation equals regulatory capture equals concentration of power, but I always found this to be an overly simplified picture.” On messaging, he said, “I do not agree, and my messaging has been disproportionately negative. It has been about equally balanced between risks and benefits.” He took a little responsibility: “By far the most accurate criticism of AI companies, including Anthropic, is that we haven't yet delivered on our big promises to benefit the world. That is totally on us.”

The most quoted line was, “The thing that will work is actually curing cancer, not glitzy marketing questions.” Sacks, you've been talking about Anthropic for a year and a half. He never responded, and then Gavin comes out and he suddenly is very talkative and responds. What do you take from the fact that he never responded to you—and then he responded to Gavin? Then we'll get into the details.

1. Dario Answers His Critics

Well, in fairness, JCal, he did respond to me.

He did?

I don't want to center myself, but the reality is he did not respond to what Gavin said. He easily could have said that what Gavin said he said is untrue, like Sholto did. He did not weigh in on that issue. Instead, he really weighed in on the regulatory capture issue.

That's the thing that I think drew him out here. I think he is a little bit irked by the fact that we keep accusing him of regulatory capture. This is something I've been saying since last year: They were engaged in a sophisticated regulatory capture scheme based on fearmongering, and those are the points he's really responding to. I think the thing that he doesn't like about it is that he feels like it impugns his motives, that he's not pure and pursuing all these regulations—

Curing cancer.

—for the right reasons.

Yeah.

My response to that is, look, there's no question that Anthropic has been extremely aggressive about seeking to implement its preferred regulatory frameworks at both the state and federal level. That is regulatory capture. I think that Dario may be sincere in his beliefs. I'm not saying he's doing this for pecuniary reasons. Nonetheless, he is seeking to capture the machinery of the state on behalf of his political agenda. There's just not much to debate there, I don't think.

There's a few other things to say about this, so let me go to point number two first. He says that he's not responsible for the negativity around AI, that he's been balanced in his portrayal of AI. Oh, really? Do you expect us to have amnesia about the fact that just about a year and a half ago you said very specifically that within 1 to 5 years, 50% of entry-level knowledge workers would lose their jobs?

This was not just some idle speculation on a podcast. This was something that they engineered and amplified into a full-fledged media campaign. You may recall that President Obama even retweeted this, and many other Democratic Party officials did the same thing.

Moreover, this pattern repeats itself with the whole Anthropic blackmail study. Remember, he's got a team at the company. They're supposedly studying alignment. They create this highly contrived study where they prompt this model over 200 times until they finally get the headline-grabbing result that they wanted.

Again, my point here is that this is not someone riffing on a podcast about the alignment risk of AI. This is them actually creating a study, engineering that study, and then pumping it to create the headlines they wanted. Dario, his sister, who's the president of the company, and other executives went on 60 Minutes and breathlessly pumped this thing.

Since that study came out, a lot of people have looked at it and said, “Look, this thing is highly contrived,” including the UK AI Safety Institute, which basically said that this thing was created under highly pressurized conditions and that they engineered the result.

These guys have created campaigns to make people fear AI. Why is the public so afraid of AI? They think it's going to take their jobs. They think it's become the Terminator. Those are exactly the fears that Anthropic has put into the media bloodstream, and no company—and therefore, you'd have to say, no founder or CEO—has done more to promote, pump, and amplify these fears than Dario has. The fact that he can't see that either makes him, I'd say, disingenuous or delusional.

I'm going with the latter.

So that's point number one.

Yeah.

I really think that almost everybody who read that and has been paying attention felt the same way. It's like, Dario, you really have a blind spot with respect to your own behavior.

By the way, can we talk about that one for a second, or no?

Yeah, let's do it.

2. Doomerism Hits Data Centers

Just to build on what you're saying, the issue that we have right now is actually pretty problematic. If I had to take 3 different vectors of actions and put them together, you've had a very disjointed, disorganized, and frankly pearl-clutching, disingenuous AI-doom narrative. It's actually seeded, and now it's growing.

As a result, what's happening? I was really shocked when I saw Governor Abbott of Texas put down this executive order around Texas data centers and limiting their growth. Josh Shapiro in Pennsylvania just did the same thing. Axios reported that the GOP sent out a memo to the executives at the various AI companies essentially saying, “Hey, stop rage-baiting people, because we're about to lose this critical GOP Senate race in Ohio largely around this issue.”

So you have doomerism, and you now have mainstream political pushback from both sides of the aisle, which I think is extremely dangerous. The third leg of the stool is rising yields. By the way, the Treasury Department just doubled its bond buying.

If you put all of these things together, as a casual observer, what I would say is that we have made some really stupid, unforced errors in the United States. The coda to all of this is that, if you put all of these things together, rising yields constrain the money supply in terms of how people want to invest in risk-seeking assets because the risk-free rate just keeps going up.

That causes people to be less likely to invest in all the data centers. Data centers themselves are being shut down. Where does that leave frontier model companies? They're the most at risk. Why? Because they're not investment grade, and they're in the worst position in terms of their balance sheets. They're not in a position to get access to the compute they need to be successful.

Even as you grow, you're actually making your problem worse in the sense that you need more and more capacity to sustain that growth, which you can't fund because all the negativity has turned people against you. All roads lead to a big question mark and a head-scratching moment for me, where I'm like, “How did we end up in this situation?”

Whether it was accidental or whether it was well-intentioned risk management, whatever it was, we are now in a very precarious situation that we could have frankly avoided.

Totally agree.

Friedberg, your thoughts on this anti-data-center movement or on Dario Amodei going direct and writing the double missive? Do you think he's sincere, delusional, or some combination?

Why don't we steelman it?

Yeah.

3. Frontier Labs Fear Misuse

If we put ourselves in the shoes of a frontier lab leader, the leader is seeing capabilities of models that perhaps we don't see because we're working day to day in enterprise settings, making stuff and doing things in a benign, non-nefarious way. They're sitting there and pressure-testing models, and they figure out that there are ways you can jailbreak models to get them to do things that they're not supposed to do.

You can use them to design bioweapons, make viruses, and make infectious diseases that don't exist today. You could use them to design cyberweapons, get them to do things online, break into bank accounts, take people's money, and all these sorts of things.

You can prompt them to do things that could broadly influence society by creating fake media, fake videos, and fake images, propagating them through social media, and hacking the algorithms. X's algorithm is public, so theoretically an agent could read it, copy it, understand it, and figure out a way to manipulate the broad public.

You realize that, as you're building these tools where there are a lot of benign, positive, world-leading, progressive applications, you see all the bad stuff that could happen with them. You can keep trying to build systems and ways to stop them from doing that. You can build all your own code that reviews every prompt that comes in and every API call. Yet somehow, a network of agents can be used to do things, and they can start to do things that we don't want them to do.

And so let's assume that you're sitting there and you're the executive of the company building this great technology. You believe fully in its capabilities and its benefits to society, its benefits to the economy. But you can't get a grip on how to keep it under control and how to stop social engineering—

Okay, wait. Hold on.

Bioweapons and cyber weapons. What do you do differently?

Okay. You're going down an exceptional path. I love this. There is one thing that you must do if you believe that this is true, to further steelman this thing. You cannot obfuscate the thinking tokens of a model, then. What do I mean?

If you just compare open-source models versus the closed frontier models, there is no way for us to understand what the thinking tokens are. You give it a prompt, you get back something, but all the stuff in the middle is obfuscated. Claude obfuscates it, okay? So do the OpenAI models.

I would say, if I was really afraid, the first thing I would do is try to find a way to unobfuscate these thinking tokens so that you could have a third party actually legitimize what you're saying. What's interesting to me is that I haven't seen the same doomerism from any open-source model company, including NVIDIA.

The thing with the open-source models is that you can actually see it thinking, so you can see its misalignment in real time. Instead, we have to basically agree to their interpretation of tokens that we can't see, that only they can see, and that they can interpret however they want.

Without judging whether it's right or wrong, I would say, give them the benefit of the doubt that they're trying to do the right thing. Make the bloody thing available to other people, then. Let somebody else with some amount of judgment sit in the room and steelman it with you.

4. AI Regulation Splits The Panel

So then let me make this point. Sacks, there was this argument that Demis put out for a FINRA-style regulatory system, where industry participants that understand technology, move quickly, and are aligned with commercialization actively review each other's systems and each other's models via an industry consortium that hires and assigns scientists to work on this stuff.

They're just doing it, versus having it be owned and managed by the government and creating some regulatory-capture scheme. But then this week, you were pretty against the FINRA system in your tweets. You seemed to be supportive of it the first time around, when Demis originally published it.

It seems to me that the proposal from Demis that Dario, Sam, and Elon all signed up to, you're now against. But that could be—

Well—

—a solution to, if there is—

No—

—genuine concern here, given Chamath's points—

Let me clarify that.

—that the industry can kind of police each other.

Let me clarify that. I think designing a new regulatory agency for AI that's designed after FINRA, which is a pretty sleepy regulatory agency that writes rules for stockbrokers in a fairly mature industry that's not dynamic at all, is a horrible idea.

By the way, FINRA is widely perceived as just protecting incumbents. Talk to any startup founder who's had to deal with them. They view FINRA as existing to protect the big banks from disruption. It's just not a good model to base regulation after.

Well, now, just real quick—perhaps the NFA, the National Futures Association, which is also one of these self-regulatory organizations in the financial markets, is a better alternative, because that's a pretty dynamic market.

Okay. So now what you're saying is very interesting, which is you mention the term self-regulatory organization, or SRO, and there are many different models. One of them may work, and that's been my point.

The one I think is the most appropriate for this circumstance would be something like the MPAA, which does not report to the government at all and was formulated by the motion-picture industry in response to the Hays Code.

What the government said is, “Get your act together, guys, or we're gonna have to do it for you.” What the MPAA did then was promote standards that forestalled more intrusive, heavy-handed government action. So that is basically the model that I said I could get behind.

But is that the model that Dario supports? No. Dario, and I think Demis, want a FINRA, which is basically this rulemaking regulatory agency that would not only promulgate standards but also do pre-release model testing, okay?

It would then report to the government and officially be enshrined in law. The reason why everyone says, “Well, Elon's behind it...” I've talked to Elon about this. Elon does not support a FINRA for AI. What he could get behind is an MPAA for AI.

And that's my point: Demis wrote something very high-level—

And by the way, that's what I've been saying on this podcast, Sacks, since the beginning. I have been saying, “MPAA: self-regulate.”

I get it. But look, here's my point. Demis wrote something very high-level, and so when you say something like, “Well, everyone supports Demis' idea,” there are many versions of Demis' idea.

Yes, I hear you, and I think—

Some of them could be okay, and some of them are bad. What I'm saying is, if it's a FINRA model, that will not work.

Yeah. Having spoken to Demis at length about this, I think the general framework is an SRO in the right structure. However, FINRA is often used as the example of an SRO simply because it is the most well-known SRO.

You are correct: I think there's a mechanism by which this thing needs to operate that is distinct from any other SRO. But the idea is that the industry gets together, and to Chamath's point, they get to look at each other's models, they get to police them, they get to have independence, but they don't get to be controlled by the government. That—

Well, here's what I think is important—

—might solve the problem.

Just to be clear, this FINRA-for-AI idea—Dario's behind it, Demis is behind it, they—

Just call it an SRO for AI. SRO. Yeah.

It's not an SRO because it's not self-regulating. It's a regulatory organization, and the S part is a total fig leaf designed to hide the fact that this is a new regulatory agency that reports to the government, that does pre-release testing and approval of models.

I call it a DMV for AI, because I think what's gonna happen—

Hmm.

—is all these models are gonna get lined up in a queue, waiting to get their test done and then released, and it's gonna slow us down horribly. Have you ever been to the DMV waiting in this queue?

It's gonna be like the App Store.

It's a terrible idea, and it's just a repackaging of what Dario has always wanted, which is an FDA for AI. Now he's saying it's an FAA for AI. You know how long the FAA and the FDA take to approve new products?

Yeah, we mentioned it last week.

Years.

Yeah.

The FAA takes—

Yeah, we'll lose to China. China has the advantage. Yeah, exactly.

China's gonna win if we do this.

But Sacks, here's what you should look at: China's rocket that they landed yesterday. You know, it's completely steel. And there's a good analysis that was done on how quickly they were able to do it from the last time they had an explosion of this. Then they got full steel done, and Elon's still waiting for FAA approval on his steel Starship.

You could actually watch it playing out in China, where their regulatory agencies are allowing them to move much more quickly. As a result, they're starting to gain real traction in the space race.

Separately, a lot of biotech has moved offshore from the U.S. for the same reason that you're pointing out, which is the FDA approval cycle is so long, and the stages slow people down so much. China has completely revamped its rules for getting to clinicals, getting to phase 3, and then completing them, so all the biotech investing in the world is now moving to China.

Totally.

The fact that Dario could want to, and has called for, including in very recent blogs, a regulatory apparatus modeled after the FAA, and that that is the right model for AI, in my view, is completely discrediting. And yes, we will get smoked by China, because they will not subject all their companies to these constraints. Now, hold that thought.

Okay, so let's play this out. Let's play this out.

So—

Yeah.

Could we go back to your steelman for a second?

Yeah. Yep.

Just because I believe that we should not have a DMV for AI does not mean that I'm against AI safety. So let's talk about some of the things that can be done.

First of all, I think it's good that these frontier labs take safety seriously. Sam just came out with a post, I believe yesterday, saying that they're actually gonna slow down the release of some of their frontier models because they felt that their safety stack was not keeping up with their capabilities. Okay, that's responsible.

By the way, do you know why I think they're doing that? Because they have huge and ample liability to take responsible steps. If they don't, they're gonna be subjected to all sorts of product-liability lawsuits.

Look at what's happening to Meta right now. They're getting sued because supposedly their product led to social-media addiction, and they're being sued for over $1 trillion right now. This comes on the heels of them losing a $6 million jury trial in Los Angeles, where a plaintiff sued them for creating her body-image issues.

So we are seeing massive product liability lawsuits against big tech companies like Meta and Google, and if OpenAI and Anthropic are smart—and I think they are—they do not want to be next. These companies have ample reasons to seek liability protection, not to just race out their product. Safety is part of what they have to get right.

Now, how should the industry coordinate with each other? There are tried-and-true ways of doing this. Companies in the industry publish papers, technical papers, and blogs. These are good ways of sharing best practices. They attend symposia and conferences, and they present at these things. They do requests for comments as part of open standard-setting exercises.

There are ways of the industry collaborating on safety in very productive ways, but what is the common denominator of everything I just said? They are all open. They are not in secret, and that makes those judgments contestable, which gets to the point Chamath made earlier: We need to see the thinking behind this.

What I'm against is these companies meeting in secret, whether it's part of some sort of FINRA or otherwise, to set standards that are not transparent and are uncontestable. You know what's going to come out of that? Regulatory capture and anticompetitive behavior.

You can't—

So there are ways of doing this that will protect the public, that are open and transparent. That is what we should lead toward.

You will not see it happen, and I'll tell you why. Again, the technical details matter. If all of a sudden competitor A, B, and C could see every single trace and span that your model generated as an artifact of going from input tokens to output tokens, you're essentially giving away a lot of your recipe. I just don't think that if Anthropic isn't willing, even if I'm an API customer, to share my thinking tokens from my company with me, they're sure as hell not going to share them with OpenAI and anybody else in a room.

Okay. Let's just do one more steelman.

Not going to happen.

So let's say that Dario's successful. And, JK, I'll ask you the question because I've taken over as moderator of this segment.

Go ahead. No, I mean, this is Sacks's wheelhouse. I wanted to hear his whole position on it.

Okay. Let's just say that Dario's successful. There's some regulatory scheme that's created that does, in fact, to Sacks's point, slow down the industry, slow down model development, and China's frontier labs make significant progress and start to leap ahead. They get this RSI, this recursive self-improvement, going, and their models become the best in the world. They only get better and better and better.

JK, we'll start with you.

Okay.

And then we'll go around the horn.

Okay, world's greatest moderator.

What is the implication for the American worker, the American business, and the American consumer in a world where China just has the best AI and all the data centers? What happens in our world? What is the consequence? Everyone keeps hearing, “Oh, China will win. China will win.” Who cares? What does that actually mean?

5. AI Backlash Is About Jobs

I don't think China's going to win. I think it'll be—there's not going to be one person who takes the whole prize. I think it's going to be incremental improvement, so I don't think that matters.

I've been thinking about why this backlash is happening for the data centers, actually, and it was kind of shocking to see Abbott come out and Shapiro come out because those 2 people were the most pro-data centers, right? If you double-click on it, they just want to make sure you're not tapping into the grid, and the water concern, which we've talked about here, is not too serious. But I think it was very performative, and I think you and I were talking about this on the back channel: They both are doing this because their constituents, the people who vote for them, really care about these data centers and they want to stop them.

So then you have to work backward. Why do they want to stop data centers so much? Why are Americans so opposed to AI right now? The perception of AI was already pre-baked with science fiction, of course, so we needed to work to get past the Skynet-Terminator vision. But Dario told everybody, “We're going to lose our jobs.” Elon said, “You're not going to have to work.” It was a more world-positive view, but he did say that.

Sam Altman did studies on universal basic income, and he told everybody, “You're not going to have to work.” UBI. Saying that kind of stuff to people who have to work for a living, who have no savings, and who are watching, administration after administration, not be able to tame inflation—whether they're Democrats or Republicans—and then watching the minimum wage just stay the way it is and barely move while the stock market hits records, and then all of us, all of us being tech executives, get super-rich, you wonder why people are now embracing socialism.

You wonder why you have Abbott and Shapiro saying, “You know what?” And Ro Khanna, worst of all, saying, “I have to be anti-capitalistic. I have to be a decel. I have to decelerate.” It's because they were supposed to see some relief, and they haven't.

And then what do people do who do not feel they have any power? They look to find power, and if you look at what's happened, they are sharpening the guillotines. That metaphor should wake people up, whether it's the UnitedHealthcare CEO getting killed or people firebombing Sam Altman's house.

We need to wake up and look at why people are so upset: It's because they can't survive. They're having a hard time with affordability and their wages. And what 2 things can they do? They can vote for socialists, and they can actually stop data centers, and that's what this is about.

Nobody—no regular rank-and-file American—cares about AI safety. That's not the discussion for them.

Yeah.

The discussion for them is, “Why am I not getting rich and everybody else is?”

I would tune your argument by 1 degree. I really think you're onto something. But the 1-degree shift is that I don't think they think about it entirely that way. I actually think they think about it as: If I think about who has won over the last decade, this generation of tech oligarchs who have won are pretty displeasing. They're not aspirational, and they're not the people that one would look up to. They're not Steve Jobs by any stretch of the imagination.

And now they look at this next generation, and they're like, “Wait, we're going to go from billionaire, decabillionaire, centibillionaire to now trillionaire, and it's those guys? Those guys are even worse than the last crop.” That's exactly what's happening.

And so I don't think it's as specific as what you're saying, Jason. I think it's vibes, and the vibes are terrible. They're looking around, and they're like, “If we can just slow this down, maybe another batch of people will show up, and they will actually mean to do the right thing and will make me at least feel better, and I won't be so begrudgingly hating the fact that—”

Okay.

—you’re about to make a trillion dollars. It is precisely that sensation, and the folks who are at the front of the show on this are not trustworthy. That's what it is.

Yeah. Well, I—

Zack, let me ask you my question.

Yeah, go ahead.

I need to hear someone answer this question.

Okay, but I'll answer your question. I'd like to.

Yeah. So my question again: If there is a regulatory slowdown with AI development in the U.S. and China races ahead, what are the implications for American business and American consumers?

They will win.

How does this actually affect people?

Oh, it's huge.

They will win. They will win.

Look, there are big—

What does that mean? What does that mean?

Well, there are big—

Help people understand. If I'm sitting at home—

Okay.

—I work in Dallas. Yeah.

Okay.

Yeah.

Let me give you a counterfactual. Imagine if China had won the internet. Imagine if all the major internet companies were Chinese right now.

Well, I can't. I need you to imagine for me. Tell me what this means. Give me a scenario of what this means.

Okay.

Yeah.

So imagine that, instead, Google and Microsoft and Meta and all these companies were actually Baidu and Alibaba and Tencent and all the rest of them. I mean, obviously that would be way worse for the United States. Our economy would not be as vibrant. It would not be as cutting-edge, and all that wealth would have gone into China.

There are also military or dual-use applications of this technology, especially with AI. The algorithms power all sorts of weapons, drones, things like that. So we'll be behind them. I mean, look, America is not going to be the number 1 country in the world economically and militarily if we lose AI. It's just that simple.

Chamath, do you agree? What's your view of the world—the U.S., business, consumer, individual—in a world where China has won, quote, “in AI”?

I don't know what we—

If you had to define—

What do you mean by “won,” Chamath?

I don't know what it means.

Okay. So what I'm trying to get to—

Yeah.

—are the implications—

If you could define that, it'd be easier to answer.

—the implications of regulatory slowdown. So if we do slow down AI, what are the implications for people? How do people understand tangibly what that means?

Unless we actually go and ban open source—I don't even know how one would implement a ban, so I think it's just a worthless exercise—I think we're going to be just fine. I think the consumer's going to win. I think American companies will win, and I think the American economy will win.

The reason why it will win is that what we are finding—I'll give you an example of a really interesting insight, and this is about closed versus open. Three years ago, all there ever were were models. You would look at some benchmark, and there would be a score, Friedberg, and you would say, “Wow, the one with the higher score is better.” In the last few years, what's emerged is what's called a harness. A harness just wraps the model. Codex, Claude Code, Hermes, or OpenClaw—these are harnesses.

You know what the interesting thing is about open-source versus closed-source models? When you take a closed-source model, it does really well on a benchmark. When you wrap it in its own harness, it decays in capability. When you take an open-source model and use any other open-source model, it improves in capability. What is that trying to tell you?

What I think we are finding out—and this data is just going to be a tsunami over the next year—is that there are all kinds of ways of using open-source technologies that are meaningfully more performant and dramatically cheaper than the closed-source alternative. Unless America stops us—us being consumers and business owners—from using the cheapest, fastest, best thing, unless they stop us from doing that, we will win. What we are debating at this point is which private American company will win or lose. I go back to the previous argument. I don't think Americans care that much whether any of these closed-source companies win or lose. I think you're seeing that in the policy.

So, Friedberg, my explicit answer to you is: as long as we don't do some draconian thing like trying to stop open source—

They are—

—the American consumer will win.

I got bad news for you, Chamath. An open-source ban is coming. They're not going to call it that. You know what they're going to say? They're going to say that we simply have to apply the same standards to open models that we apply to closed.

Here's how they do it step by step. Let me explain how regulatory capture actually works. First of all, you have to get this regulatory apparatus. Dario wants an FDA for AI, but he doesn't have enough political support for that, so instead they do this Trojan horse of a FINRA for AI. They call it self-regulatory. It's not really. But anyway, that gets them off the ground.

Now they've created the standard-setting organization. Now they've got pre-release model testing. Then the pressure grows to codify that in law, so that happens next. And then what they do is they say, “Look, all these standards need to apply equally to all models.” But here's the problem with that: open models and closed models, as you know, are technologically different.

Go back to Dario's Senate testimony in, I think, 2023, where he said that open models, if sufficiently advanced, are dangerous. Why are they dangerous? Because they cannot be centrally monitored and controlled and rolled back and withdrawn—

No, but Sacks, it'll cripple the American economy.

Hold on, unlike a closed service like the one he hosts. Okay.

I understand, David, but—

No, hold on. Let me just finish the point, then you can—

—it'll cripple the American economy.

Please, let me just finish the point, then you can—

Yeah.

—you can respond.

Dario basically says that open models are dangerous because of these characteristics they have, and those characteristics are immutable. Technologically, they can't be changed. Once you release an open model into the world, you can't roll it back, and you can't monitor exactly how people are using it because they run it on their own hardware. Dario says this is what makes open models dangerous.

So what they're going to do is they're going to have the standard-setting body say, “Well, we have to set the standards for AI safety.” By the way, Dario and OpenAI, they're going to fund the whole thing. They're going to contribute all the compute. They're going to be behind it. They're going to be the ones coordinating with the government officials because, frankly, people in government have no idea how to monitor and control and set standards for AI safety. Technologically, this is way beyond them.

So they're going to go to these companies and say, “Tell us how to do it.” And so what will happen is the standards will get set, and then it'll be a very simple matter of fairness to say that the standards need to apply to open as well as closed models. The open models cannot comply in the same way, and gradually they will be shut out of the marketplace.

Shyam, if you want to add to that?

I understand. I understand, but the American economy would be crippled overnight. Why and how?

So what? You're saying regulators have never done things that are crippling?

Okay, let Shyam—let Shyam answer it. So let's take, again, the example of Coca-Cola. Coca-Cola is an extremely profitable, cash-generative, well-run business. It has existed for, I don't know, 100-plus years. At this point, it will exist for many hundreds more if left to its own devices. It is a global business. Take Pfizer—same situation: well-run business, global business.

Now what you will tell them is, unless you tell them, “Hey, anywhere you operate, you need to follow American rules.” Okay, that's one thing. Or it'll say, “Inside of America, you need to follow our rules, and other governments will have different rules.” Now, what is Coca-Cola to do? Run the slower, more costly, less effective, less valuable model in the United States? Sure, they'll do that, and their U.S. business will be crippled.

They will have a better, flourishing business abroad. They will invest abroad, and what you will see is FDI—foreign direct investment—in the United States go off a cliff. And what you will see is direct investments into other countries of the world rise. You've seen this, by the way, in China. When China made all these moves to cripple the Chinese tech companies, I think this is the best analogy.

There was a moment where China was at a fork in the road, where the large oligopolists there, the large tech oligarchs, were trying to seize power, and the tip of that spear was when Jack Ma gave that very famous speech on the eve of the IPO event. What happened? Xi Jinping cracked down. They took complete control. They said, “We are now operating together. Everybody that is worth anything will have a representative from the Chinese government on their board. We have the golden vote. We're going to work together.”

When you look at what happened to foreign direct investment and you look at what happened to their economy, it fell off a cliff. That will happen here if this happens. So they can try, and everybody will say, “Okay.” And what the Coke CEO will do is redirect all of his capital allocation to outside of America. So will Pfizer. So will I. So will anybody.

And then what are you going to do, take away our passports so we can't travel? So I think it will be a very visible way of just destroying the United States, and you can do that if you want to, but you will see it play out almost instantaneously in the equity markets and in the capital allocation of companies.

I agree with you about all the negative consequences if we ban open models, but that doesn't mean it can't happen. Look at all the negative consequences of banning data centers. Data centers are going to be 3% of GDP. It's basically creating all the gains in the stock market—

But they're not voting against that. They're voting against 2 people that they deeply dislike becoming trillionaires. That's what they're voting against.

That's a big part of it. Chamath, what do you think?

Well, I was just going to add to this. We already know what would happen with AGI, Friedberg, because your original question of trying to explain to Americans what happens when China wins—we already know what happens when they win. Products get cheaper, and you get more access to them. If they solve cancer, what are they going to do with that solution? Are they just going to keep it for themselves? No, they're going to sell it to Americans, just like they sell phones cheaper, just like they sell cars cheaper, and we try to block them.

And then what are Americans going to do? We're going to lose our jobs to them. And so that would be the way, I think, to explain it to them. But if we want to turn this around, this really is just about convincing Americans that they're going to have jobs and that this is going to benefit them. The American people are scared of losing their livelihoods. That's the beginning and end of it.

The place where I have some expertise is in having invested in Zipline and Uber and a bunch of self-driving companies, and that's where—

I thought you were going to say being poor.

The money's not the issue here. But the thing that I think is where this is going to play out is really on the autonomy and those jobs. Zipline just did a partnership with Uber.

I’m an investor in both of those companies, obviously, and they’re going to do 1 million deliveries a day. Then you look at the massive progress by Waymo. They’re doubling the number of cars on the road every 6 months or so. Tesla’s about to go Cybercab.

What people are perceiving is that, in their community, you’re not going to have any money staying in their community. Because those drivers—as much hand-wringing as has occurred—the money at least stayed in the community. They weren’t making enough; they were only making $18, $16, $20, $21 an hour, whatever it is. But now we’re taking that away from them.

And then where does that money go? Where does the money for a robotaxi, a Waymo ride, or Zipline go? It goes up into the cloud and then comes back down into Silicon Valley, polarizing more wealth. And that’s where you’re going to see it. I will predict it now: they’re going to ban Waymos in New York, Boston, and Washington, D.C., and they’re going to limit them with licenses.

We must figure out how to explain to Americans that we’re going to raise the minimum wage. When I say we should raise the minimum wage and focus on that, people are like, “That’s crazy.” But when we talk about an experimental tax cut, people are like, “Oh, yeah, we do an experimental tax cut. We’ll change the cost basis of capital gains. Great. Let’s try and see what happens.”

Nobody wants to try to raise the minimum wage. Nobody wants to give people an explanation of where their job will be in the future. And that’s where—whoever’s making the point, I think it was you, Chamath—about who are the spokespeople here? We need spokespeople who are saying, “This technology will empower you to get a great job.”

Paradoxically, the guy who did the best job explaining this—or the 2 guys—were Jensen, obviously, who said, “You’re not going to get replaced by AI. It’s going to be somebody who works with AI,” and then Zuckerberg, who said, “I’m going to give you a bunch of free tools.” In his manifesto—going direct—he was like, “I’m going to give you free tools. I’m going to empower you. You’re going to get agents. Your life’s going to get better.” Dario and Sam have not done that; they’ve done the exact opposite.

Meta actually did a really great thing. Zuck did a really great thing, which is he created this new program that’s almost like a training school—

Yes.

—for people to go and learn how to get these various kinds of jobs. It’s like a trade school—

True Wealth Academy.

—for laying fiber and doing these new blue-collar construction jobs, which we have a huge shortage of. That’s why it’s hard to build these data centers.

Yes.

They’ll pay your salary or something. It’s a great program.

Yeah, so it’s—

Yeah, we need more of that. So how is Zuckerberg all of a sudden the best spokesperson for our industry?

It’s a very low bar.

I guess that’s it.

But look, this all goes back to the fact that Dario—not exclusively, but again, he was the one who put the most specific numbers on it and therefore made it the most tantalizing and salacious for the media.

Fifty percent of entry-level knowledge workers are going to lose their jobs, not in the distant future, but as soon as 1 year from now. We’re now past that 1-year mark, by the way. Has he updated? Has he said, “I was wrong”? Not really. Sam did, by the way, to his credit.

Mm.

Well, why don’t they just talk about jobs? Sam Altman and Dario should start talking about jobs.

What jobs?

Sam actually came out and said, “I was too negative. I was wrong.”

He did, yes.

Dario has doubled down. As far as I know, he’s never recanted. He’s an ideologue. But to your whole point, yes, I agree with you that the animus and anger toward data centers flows out of the fears about jobs, as well as other fears about AI. But it’s the job loss that is the main concern, and there is no evidence of that. We’re now years into the AI boom, and there’s no evidence of that. We only see—

It’s the opposite.

Net job creation.

Exactly.

Yeah.

It’s only been the opposite.

I had massive concerns about it, and the business opportunity is so great that I’ve got 7 job openings. So I keep saying—

I will also say something else.

I’ll just say—hold on.

When we all think this way about it—

Hold on, hold on. This is important.

When we all—

Hold on. JCal’s got an important announcement here.

Listen, what I’ve discovered is that if you can convince your team to embrace the tools, so many business opportunities emerge that you wind up hiring people to work for the people in your organization who embrace the tools. So we have become so fast at building the syndicate and building our programs that I’m like, “You know what? We need to add a couple people.” I thought we were going to automate them away—

I’ll just say—hold on.

—and we’re adding.

I’ll just say another general point. Silicon Valley has completely lost what made it great. It was a place for weirdos, for people at the fringes, for people who really were idealistic, and it has become something that’s far from that now. It is the equivalent of what Harvard used to be.

It’s a credentialing place. It’s a place you come to say, “Oh, I came to Silicon Valley. I worked at a startup. I got into YC.” When we came, we used to aspire to meet the Bill Campbells of the world, Mike Moritzes, and John Doerrs. Those folks are gone. We used to aspire to work for Steve Jobs. He’s not here.

And in that breach, all we have done is focus on making money in every single way that we can, hand over fist. I think there’s nothing wrong with that, except now is the point where we need to do a better job of acknowledging it and trying to do a little bit better. Every time we don’t—and we’re given a chance to be better and we’re not—the hatred builds. The animus, as you said, David, builds, and that’s what we’re seeing now.

Yeah, there’s no question about that, but can I defend Dario? Let me just say something in his defense.

I’m sorry. My headphone went out.

You said you want to defend Dario? Breaking news.

I actually think he is very idealistic. I don’t think money is his main motivation. It’s just: what are his ideals? He wants to bring forth this new superintelligence—machines of loving grace—that are going to apportion all the resources of society while humans twiddle their thumbs.

It’s not a vision that the average person believes in or is attracted to. In fact, it’s dystopian, and that’s why a lot of their messaging is dystopian. Because of that vision, no matter how idealistic his motives, it is why he does not inspire trust.

I don’t think he’s lying to us, and I don’t think what he’s really trying to do here is capture all the wealth for himself. I actually think he’s—

Well, it’s quite the opposite.

He’s trying to bring forth this future, but it’s not a future that I think the American public wants. And, by the way, he should have to defend that. If you think you are bringing about the singularity—and let’s face it, he probably is in the lead to do that, to bring about recursive self-improvement—you should defend this future.

I see all these people simping on X who are like, “Oh my God, Dario, thank you for coming forth. Thank you for coming down from Mount Olympus and slumming it with the rest of us on X and finally engaging on social media.” And it’s like, no, he should have to do that. If you’re going to put yourself in the lead to bring about this crazy future, you should tell us why you’re doing that.

Well, and he should speak more. I think this is a key point. To your point that it’s not about money for him, the 7 co-founders of Anthropic—this is my understanding, and it’s been documented, I think, in a bunch of press reports, but they never talk about it—donated 80% of their shares. So 10% of Anthropic is in some sort of trust, which, at $2 trillion or $3 trillion, 10% of that is going to be the largest—

Well, it’s going to be the largest Democratic PAC. It’s going to be the largest PAC in the world.

Now, how will they use it? But why aren’t they talking about how they’re going to—

DSA.

What could go wrong?

The largest DSA PAC.

A trillion dollars to EA causes. What could go wrong?

Well, let’s hope they put it toward something that actually is humanity-positive. They all own only about 1.8% each.

That’s a term of art.

Yeah. And then I do think going direct is critically important. I think we can move on after this.

What have you got to learn? You’ve got an AI to talk to.

No, no, no. The playbook is very simple. Number 1, you’ve got to write it all yourself, which he clearly did.

Are you sure about that? I don’t know. It felt very Claude-like to me.

I think so.

It felt very Claude-like to me.

It felt like it was him.

It felt very Claude-like to me.

You think? You know the Claude voice? He is the father of Claude.

That's true, actually. Claude could sound like him rather than vice versa.

It could be the other way with causation and correlation.

Yeah.

But he did the exact playbook: you have to exhaust the audience. You have to go thousands of words, like Zuck did, and then you have to get ahead of problems. That's where the leadership is breaking down. You have to talk about what's coming next, and then never, ever apologize or give an inch. Just attack whoever's attacking you: fake news.

Right now, I don't think he can win the argument by being transparent, because what he's selling is fear—

Hm.

—in order to motivate the government to give him control to bring about this dystopian future that he believes in.

Yeah.

At least that's my distillation of his approach.

Just do what Zuck did. Create a program for tool-belt generation. Come up with ideas like that. Do more of that. Friedberg, you get the last word. Go.

6. Recursive Self Improvement Takes Off

I think that it's worth having a view on whether or not you believe in recursive self-improvement. Recursive self-improvement is this theory that, at some point, an AI model can spin up a whole bunch of agents to do a bunch of work, and they can work together to make a new AI model, and the new AI model is better than the last AI model. At some point, this all becomes effectively fully automated and self-evolutionary. The AI is now improving itself continuously, and that is this idea that you have the singularity, as some people would call it, or you have ASI, artificial superintelligence, that arises from this capacity. But the capacity is recursive self-improvement, RSI.

Anthropic has stated they're not sure if it's going to happen, but it could, and they're working toward it. They expect that there's a path to get there. Having looked at this and thought about it, I happen to believe that it is possible. I think it is likely to happen soon. People I've talked to in the industry do think it's real, and there is this kind of philosophical debate because companies have a different view on the path forward, where it really needs to be human-constructed and human-designed. There isn't this idea of, "Let it all go, let humans detach from the AI, and let the AI evolve."

This is really like the movie Her, where the thing just takes off and evolves. But if it does happen, and if RSI is a real construct, then you're almost fighting—it's like a fool's errand. It's a fool's errand in the sense that all this idea that you've got human intervention in the stages of model development—

Oh, you're scaring people again. You just—

Well, no. No, this is my point. If RSI is a thing—and, by the way, RSI doesn't need to lead to some dystopian outcome. It just means that the models improve faster and faster—

Without humans.

—and without humans necessarily constructing, architecting, and designing the next evolution. Then you don't necessarily have space for, "Oh, we just did one big model run every 6 months. We're going to go to the government regulator, or we're going to go somewhere and get it approved." There's just this continuous policing and monitoring service that needs to take place at some point.

The truth is that, to pull this off, you don't need governments in the US to say yes, or governments in China to say yes, or governments anywhere to say yes.

I know what you're going to say. You need satellites in space.

You need some sovereignty, some space where you can. I don't know if it's in space. I don't know if it's with Bology[?] and the deal he just did in Kazakhstan, and you've got some unique geography.

Nice.

It could be in Iceland. It could be—

In the ocean. People are floating it.

It could be in the ocean.

Seatopia.

But the point is, if RSI is a thing, you don't have a human process where you're stepping in to run this thing, so you don't necessarily have a cohort of frontier lab scientists running it. Theoretically, an individual that's smart enough, or a group of individuals that's smart enough, can go spin up an RSI factory anywhere in the world or anywhere in space.

Your attempts to ban the data centers, or the attempts to ban or to create a regulatory body—that's why I call it a fool's errand. The thing with the open internet and the thing with compute is that all it needs is the chips, the power, and a connection. If it's got the chips, the power, and a communication connection, it can run.

I'm not painting a dystopian view. I'm just saying, practically, pragmatically, the labs are all talking about RSI. If RSI is where this goes, then the labs can go and set up data centers in Iceland, in China, or wherever.

Whoever will give them the freedom is a great point.

Wouldn't we be better off having systems where we actually have those labs based, where we have those data centers based in the US, where we have jurisdiction over them, and where we have the capacity to monitor, control, and police them, rather than just have them sprung up in some random country or some random territory? So I guess my point is, we should be attracting and retaining the frontier labs, not trying to—

Yes.

—you know, chase them off.

And then there's the famous Wait But Why image of human progress going on at a nice clip, and then, boop, recursive self-improvement, and it just goes out of control, and—

I'll tell you why I think RSI is a legitimate concept.

How close are we to it? I think that's what people would want to know. How close do people think—

Honestly, time—there's a massive problem with time dilation in AI model development. What I mean by that is, we would typically consider some sort of thing that takes a linear amount of time: there's a certain amount of compute to get somewhere. There are changes that happen that are non-predictable, where you don't actually know how much time the next phase or the next step change takes. The time scales change a lot in AI model development, and that's what makes it so crazy and so hard to predict.

That's why no one's really good at predicting when AGI is happening or when the next breakthrough is happening. Then, all of a sudden, everyone's like, "Oh my God, we're in the singularity. They just solved 8 unproven math theorems in a week off of some model that's barely cutting edge that's now going to be the second cutting-edge model." So it's moving very quickly.

But a lot of the models are this mixture-of-experts concept, where you've got these smaller models that call each other to do things, and then you wrap it up and call it a big model. Those smaller models, it seems, over time, look a lot more like what you see in biology, where you have genes. A gene makes a protein, and proteins interact in really complicated ways that we don't fully grok. Networks of proteins do things, and then they complete tasks.

I think the same evolution is likely to happen with the models, where they break down into smaller models. That makes them so much more efficient, and that makes them so much more performant. We're seeing a lot of this in how a lot of the models are evolving. We call it a model, but it's actually many models underneath it that work together.

If the models actually do evolve this way, then RSI is the perfect system for figuring out the right path: How many sub-models are there? What do each sub-model's functions look like? How do they call each other? That evolution is likely going to happen without a human architect behind it. It's going to be developed through a test-and-deploy system, and it's going to be fully automated.

That's why this approach to solving that problem ends up becoming the future of how model development takes off, which is why I think RSI is probably the future. When that happens, again, it's just wherever the data center and the power are.

I remember a few years ago, everyone was like, "AGI is imminent, AGI is imminent," and then where is it? Maybe we're already there.

Could be like Y2K.

Okay, fine, but—

Could be like Y2K.

Did it turn out to be the Terminator?

Right. No, that would be superintelligence. Artificial general intelligence is like as smart as any human. Superintelligence is like you're getting towards what—

No, I get it. Look, I—

—Friedberg is talking about. That's far off.

Listen, do I think RSI could possibly happen? Yes. I think there are lots of black swans that could happen. It's very hard to assess the probability of those things. What I do know is, like you're saying, I don't think some new government regulatory agency will stop it if that's a thing, because it won't be technically sophisticated enough.

All it will do, actually, is concentrate power around that thing. It'll limit the options. When it comes to AI—let me just talk about AGI for a second.

We just want to have that technology in this country, is my point.

For sure.

And I just want to go back to the point I was asking earlier. Why do we not want China to win? Because we want the best labs. We want the first breakthrough system to be here, not to be there.

Dario made the argument in the context of trying to ban the export of chips, which is one of his hobby horses, as Anthropic has really been behind a lot of the political activity to restrict NVIDIA's sales internationally. The argument was that we want to have a data center full of geniuses in the U.S., and that eventually, what AI is going to be used for is creating all these agents that are basically Nobel Prize-winning-level scientists. If we have thousands or millions of those and China has only dozens or 100, whatever, then we're going to win everything. That was the whole argument.

But now we're basically banning the data centers. We're not going to have those data centers full of geniuses, and he's done his best to restrict the sale of chips to our allies as well. So I guess our allies aren't going to have the data centers either. We are totally shooting ourselves in the foot. It would probably be a good thing if Dario really does believe in the data centers-full-of-geniuses idea that he was arguing for last year. He should come forward and say, "We're making a big mistake by overregulating data centers. Here's how we should do it." I haven't heard him step up in that way, and it makes you question: do you only fight for the doomer side of the argument, or are there versions of the positive side that you're willing to fight for politically?

Now, Friedberg, to your 1 last point about RSI, I think this is really important. Again, it's a black swan, and it's very hard to say if it's going to happen. I do think that we're headed for AGI, or already in AGI, or whatever. But the most important thing about AGI is that we have choice. I'd much rather have multiple AGIs than just 1, right?

For example, Elon says that the way to keep AI or AGI aligned is to make it maximally truth-seeking. I kind of like that, right? That's a way to prevent it from becoming woke. What Anthropic says is that the way to keep it aligned is to have this Claude constitution that's full of liberal values, and you train it on that. They've got this EA ethicist or whatever who's writing these things. So there are different approaches. I want all of them to produce sophisticated AI, to keep checks and balances, to keep these things in check. What I don't want is just a Skynet or, you know, a woke Skynet—

Right, right.

—which is sort of the worst possible outcome. If we're going to go sci-fi for a second, there was a Terminator TV show.

Yes.

That was pretty good. They stopped it.

It was pretty good. It got canceled. This was years and years ago.

Come on.

But anyway, what they figured out in the show—the plot was that the way they finally defeated Skynet was that they created a good AGI to take on Skynet. They realized that humans just couldn't do it by themselves, so they created a choice.

Maximally good.

Yeah, they trained it to be more—

Benevolent.

—benevolent toward humans. In any event, if you're going to get really sci-fi about it, I think choice has to be our hallmark. Competition has to be our hallmark. We want multiple winners here. We don't want a small number of elites guardrailing this thing.

Mm.

And that is what you will get if you head in this FAA–FINRA direction.

Breaking, Friedberg: the new nickname club just dropped. There it is. This is what you sounded like, Friedberg: Doc Brown. "Marty, Marty, your future's not predetermined. You can make your own future. You can use RSI." Doc Brown.

Next topic.

I'm not trying to be an advocate for RSI. I'm trying to be pragmatic about—

No, you've seen the future. You're back—

—if RSI is the way things are going to go, you definitely don't want to be trying to regulate AI to stall out, because someone else will win it, and then you're literally handing the keys to RSI to them because they're suddenly going to be able to get there faster. And you don't want to be on the data centers, because then they're going to be off-site.

7. Andreessen Faces Antitrust Scrutiny

All right. There was a peculiar story this week. Andreessen Horowitz is allegedly being investigated—this is all according to a Bloomberg report—by the Department of Justice for over a year under something called Section 8 of the Clayton Antitrust Act. This is a 1914 law, over 100 years old, prohibiting interlocking directorates, which I guess is old-timey English for directors, like boards of directors. The conflict seems to stem from Ben Horowitz being on Databricks' board while another partner from Andreessen Horowitz, Martin Casado, sits on Five Trains' board. I don't know if you guys have heard about this story. It was one of the top stories this week. Obviously, Andreessen and Ben Horowitz have been all in with Trump and big donors. Put that into the mix as you will. Why is this happening now?

I can tell you. An enemy of Andreessen Horowitz, jealous of their returns, says, "What are we talking about?" Do they not understand how venture works? Is it that you invest in 1 company and you can't invest in another? I don't understand. What did Bloomberg say?

They said they've been under investigation for this conflict as described between these 2 companies.

Which means an enemy reported them to the DOJ and then said to Bloomberg—

Yeah.

—they're under investigation.

Well, that's why I bring it up: it seems so odd.

Which, to me—

And then what's the timing of it—

Which, to me, likely seems—

—and the point of it?

I'll make a speculative guess: the CEO of 1 of these companies is pissed off that the other partner—

Mm.

—is on the other board and reported it to the DOJ and reported it to Bloomberg at the same time.

Interesting. All right. Well, there you have it, folks. Moving on in the docket. I think we made quick work of that one.

Let me just say, you've been seeing the distribution of Cursor and then SpaceX. Andreessen Horowitz is on fire. They are crushing it.

They've had 2 really—

Congrats to them.

—wins in the last few weeks.

I mean, my God. Congrats to them.

Did they make—

Keep going, guys.

—how many billions did they make on—

Keep going.

It's Cursor and—didn't they make a lot of money off OpenRouter?

And SpaceX originally.

Well, OpenRouter just went out—

And OpenRouter—

—for 17—

Yeah, but OpenRouter was more like a—

It's been out a few years.

OpenRouter was more like an amuse-bouche. It was 600 or 700 million. Cursor was decent. It was like an appetizer, but SpaceX was the main course. That's like 15–17 billion. Now you're talking real money.

Just by the way, even if you want to say that VC firms should not invest in competitors, the reality is that startups exist in a very dynamic environment. They frequently pivot and change their business models. Two companies that start off in totally different spaces can end up—

Of course.

—competing with each other.

Yeah.

So now these are different board members, and they can probably set up a firewall—

A firewall.

—between them.

Yes. Yeah, Chinese wall.

Very easily.

Firewall.

And I'm sorry, these are not filing companies. What confidential—

Right.

—information do any of these private businesses have?

And, by the way, Andreessen Horowitz could also just put an independent director in there and be done with it. So if this was some ticky-tacky, very important thing, I don't know why the DOJ is on this, but—

No, I think Friedberg is right. No. Because when you make a claim to the DOJ, technically, if they open a file—

Mm.

—you are technically under investigation. And so then you can report that to a buffoon reporter who doesn't know from Shinola, and they'll just repeat it.

Mm.

It means nothing. The DOJ doesn't tell you that you're under investigation, and they also don't tell you when it's closed.

Same as the SEC, yeah.

So it's just a 1-way PR ratchet for an enemy to exact some sort of chaos. AH is crushing it. God bless them. Godspeed to them.

Yeah. It is interesting that there is even this rule, but we had some fun with it, so let's keep moving.

I want you, Friedberg, to join me and Sacks and Bestie Chamath at the All-In Summit happening September 13th, 14th, and 15th in Los Angeles. Also, our lineup has been announced, and hey, guess who's headlining the party? The same guys who sell out Madison Square Garden. I know that Friedberg has a personal relationship. Go ahead, tell everybody who's going to be our opening musical guest this year.

The boys Machine Smokers, which is going to be awesome. So we'll have them at the closing party. It's going to be great, and then we're also doing another—

Wow.

... full takeover at Universal Studios. We got a whole party at Universal, which was awesome last year.

Incredible.

Uh, where Sacks and I were with—

Yes.

... Diplo on stage. JCal, you were there, too.

Uh.

It was great.

Yeah, I was on stage dancing.

Who's Diplo?

Diplo's right, the DJ.

I'm just kidding.

Diplo. Your kids probably love TikTokers, too.

Your kids, your kids love Diplo. It would blow their minds.

They should come to that concert.

8. The Midterms Become A Toss Up

All right, we know the Democrats have been favored to take the House in the midterms, which are just around the corner. Polymarket has them at 88% favorites. But Polymarket now has the Democrats at 52%, basically a toss-up.

The current balance of power is GOP 53 seats, Democrats 47, which means, I guess, the Democrats have to flip 4 of those without losing any. 35 seats are up in 2026. There are some special elections going on in Florida and Ohio, and 23 of those 35 seats are held by Republicans. Big races to watch are Alaska, Maine, North Carolina, and Ohio.

On February 25, the day before the Iran war started, the GOP was favored 60 to 41. It's been as high as 70%. Here's your Polymarket. Sacks, what do you take from this? Why are the Republicans set to get trounced in the midterms?

Well, first of all, polls are notoriously wrong in the summer before an election, and there's some data on this. So if you're going to cite polls, let's actually talk about whether political polls are even accurate at this point.

These aren't polls. This is Polymarket, just to be clear. The polls are slightly different, yeah. These are people betting.

Okay.

I get it.

But the polls are shaping public opinion right now. Political scientist Patrick Ruffini analyzed over 3,000 polls over the last 4 election cycles, every 2 years since 2018, and what he saw is that there is a huge bias. The preponderance of polls in all 4 cycles favored the Democrats, and the average polling error across all the polls was D+3.7 relative to the election outcome.

I think this polling problem has become even worse recently, and you see it where there seems to be an oversampling of young progressives in these polls. It's not just the fact that the polls favor Democrats over Republicans; they also favor the DSA Democrat over the more moderate or centrist Democrats. You've seen this in a few elections where the DSA candidate was supposed to win by a lot, and then they basically eked it out or actually lost, like in Wisconsin.

In any event, I just think that the polls are very unreliable at this time of year, and a lot of the polls are actually created as internals by campaigns to motivate their donors.

So basically, people are upset, and then they throw people out of office. That's just the trend, and that's what happened over and over again. Yeah?

No, I just think that the polls are oversampling Democrats and progressives and have a bias. That's what the data is showing, so I just wouldn't read too much into this right now.

I do think it is legitimate to say that the Democrats are a favorite to win the House. I think the Senate right now is a toss-up, and I would say that the Republicans will ultimately keep it, if I had to predict right now. I just think that when you get back from Labor Day, that's when the campaign for the midterms really starts. Republicans need to focus on the right messages in this campaign, and I think there are plenty of good ones. They just have to remind them of all the good things—

What are they? What would you say are the top 3 they should focus on?

Well, look, everyone's been over-indexing on the war, and that's been the focus of the news. Hopefully, the kinetic phase of that is over and doesn't restart. Assuming that's the case, I think the Republicans can remind the electorate of everything that's happened over the last couple of years. I think there is a good story to tell.

First of all, remind people that the border has been sealed for the last year and a half. We haven't just slowed Biden's invasion; we've stopped it completely. Encounters at the border have fallen off a cliff to the lowest level in 50 years. Illegals aren't even really trying to cross the border anymore. It's been that successful.

Second, sealing the border and deporting hundreds of thousands of the worst criminals has actually made the country much safer. Murders have dropped around 20% year over year, and the murder rate is the lowest in 70 years, so that's number 2. Third, deaths from drug overdoses have dropped 20%, again thanks to tougher border enforcement.

Fourth, Americans saw a 17% increase in tax refunds thanks to the president's no tax on tips, no tax on overtime, and no tax on Social Security. These are real increases in take-home pay as a result of those tax cuts.

Also, the tax provisions in the Big Beautiful Bill incentivize new equipment buying. We talked about the accelerated depreciation, and perhaps that's why small-business optimism just reached its highest level in nearly a year. More small businesses are saying they plan to hire in the months ahead.

I think we do have a strong economy. Prescription drugs and auto insurance costs both fell last year after increasing a lot during the Biden administration. I think prescription drug prices have had their steepest annual decline since 1963, so there are positive things to talk about there.

Everyone made a big deal of the surge in egg prices at the start of the administration, but few have bothered to note that there's now been a 39% decrease year over year in July. Look, there are many good things for Republicans to talk about. They just have to remember everything that's happened in the last 2 years.

Okay. Friedberg, you have thoughts on what is going to happen this midterm and what people are making their decisions based on?

9. Affordability Fuels Socialism

Things are expensive. People can't afford housing, healthcare, or education, and are leaning toward this DSA concept. There was a frightening statistic on young conservatives embracing socialism.

Really? Wait, the conservatives—like MAGA folks—you're saying are embracing socialist tenets?

How are they conservatives then?

Well, I guess ostensibly they're flipping, or they're changing the composition of the issues they care about in some way.

53% of self-identified conservatives under 40 support government-run grocery stores. Not Democrats—conservatives. Take that in.

The first line of this Wall Street Journal article: A July 2026 Fox News poll found that 61% of Republicans view capitalism favorably, down from 72% in 2019. While the share who feel strongly favorable has fallen from 54% to 41%, the share who say the system is rigged toward the wealthy has climbed from 30% in 2018 to 42% among Republicans.

I don't think that this is a principled line in the sand anymore. I think that the problem with unaffordability, which I would argue continuously and have argued since the first day of the show, is a function of government spending. The more government spends on education, healthcare, and housing, the more expensive those things get rather than the opposite.

They're supposed to be able to provide access to those things, but when the government intervenes in those markets, those things get more expensive, and no one wants to acknowledge or admit that in either party. I've been to D.C. enough now to know that no one actually acknowledges or cares about this point. They all think that their job is to get more money out of the government for their constituents—every single person in every party.

There are very few principled conservatives. Many of them are conservatives by title, by party affiliation, but they're not conservatives by principle in the traditional sense: small government, free markets, et cetera. I think that this is the motivating factor.

I would argue, if you take a look at the macro cycle—the macro cycle being we've printed so much money and we've got so much debt—the 30-year is now at 5.3% this week. It continues to hit new 20-year records every day. I think we're going to continue to face a problem of unaffordability, and ultimately, there is going to be some sort of big movement toward more socialist policies that we're going to see play out between now and 2028.

That's why I continue to believe someone like AOC becomes president. I think there's going to be a wave this year, and then there's going to be a bigger push in 2028, because I don't see the arithmetic or the fiscal situation changing to actually alter the affordability landscape. I just don't.

I don't know what you could do from a policy perspective, and I don't hear either party talking about actual solutions to solve this affordability problem between now and 2028. The other big problem that I've talked about is assets. Middle-class people, people that worked in private companies, people that don't depend on Social Security or public pension funds—they've all had 401(k)s and owned homes.

So they have seen their net worth climb with—

Surge.

Asset inflation surged. So the top 50% of Americans have 178 trillion dollars of net worth, and the bottom 50% of Americans have 6 trillion dollars of net worth. It's not the billionaires. It's the whole middle class. The middle class has all the wealth in the United States, and that's because they've owned assets.

They've owned houses, and they've owned stocks. Houses and stocks have surged in value. So if you don't own a house and you don't own stocks, you were left behind. I don't see how, from a public policy perspective, we don't find ourselves walking down this very nasty, winding, dark path of socialism, because the alternative policies that could be put in place to solve these problems are not being put forth by either party right now.

So I do think that there's going to be a big socialist embrace. I've said this for a number of years, and I think we're going to see this play out in these midterms, and that's going to create the momentum going into 2028. And look, this polling data is very worrying. It's not just the Democrats. It's not just a party-based system. It's the horseshoe. It's the 63% of Americans who are living paycheck to paycheck and can't afford anything. Those are the Americans who are going to vote for this system, and it doesn't matter what party it is. They're just going to vote for it.

So, about 20 cents on the dollar goes to interest, approximately, on the federal side. Inflation—we thought we would get there. It's been 3% or 4% for the whole year. We haven't been able to get down to 2%. And the GDP concept, which we talked about here—getting to growth again, and if we can get it, we can grow our way out of it—that hasn't happened either.

Obviously, the war is a big disappointment for many folks in both parties who were promised no war. And here we are, 6 months into it. What do you think is going to happen through the midterms and then into 2028? And to Friedberg's point, is there any way to get out of this? Is this a death spiral, or is this something we can course-correct out of, in your mind?

I think that we can course-correct. I just think it takes will. I think we need corporate America to step up and do its part of the job, and we need governments, mostly, to get out of the way and not have any harebrained schemes involving new, inventive legislation that they think can solve the problem.

I don't trust any polls. Just to be clear, I don't believe this polling data. I think that we've gotten to a very dangerous point, which we already got to with the media, where the media lies. We all now know that the media lies, and the media can't be trusted. I would like to now go out on a limb. I don't think it's much of a limb, though: pollsters lie, and I don't think you can trust them either.

I think that the media acts in concert with pollsters to find the data that allows them to write the headline that hopefully allows them to get the clicks. So I would just start there: pollsters, trash; media, trash. What is really happening on the ground? I don't know, but I don't trust what this data tells me.

Now, what do I know is helping and hurting the incumbent president? I think that if we, as corporate America, get ourselves in a position to actually present the better face of capitalism, we allow the system as it stands to continue to work. And I think all of these things, if we don't wake up, are telling us that corporate America is not trustworthy anymore. Corporate America is not aspirational. Corporate America does not do the things that engender some amount of trust and respect.

Specifically in Silicon Valley, we've completely lost the script because we've gotten to the far, far, far end of that, and everybody is in it for themselves to such a degree that it's either pearl-clutching on the one hand or getting your hand caught in the cookie jar on the other. I posted about the cookie-jar thing that Anthropic did on the health care side.

Cookie jar.

Why are we in this? It's just so avoidable. Those are headwinds for the incumbent president, even if he has nothing to do with it.

What is a tailwind? I think getting out of the Iran war can be a massive tailwind. I think structurally solving energy on behalf of American citizens can be a tailwind. I think making sure that all of the long-term and medium-term investments of GDP can actually come to bear fruit can be a tailwind. All of those things—there are still a lot of levers to be pulled between now and November, and those can create pretty meaningful tailwinds.

But the headwinds, unfortunately, are pretty strong, and it requires some real coordination that the White House is going to have to do to get a bunch of these corporate CEOs in the room, to get a bunch of these Silicon Valley CEOs in the room, and to tell them, “Guys, enough of this. Play nice in the sandbox, because you're about to lose the standard operating procedure that allowed you to exist in the first place.”

Yeah.

They need to get this message clearly.

Yeah, and I can tell you the exact reason Democratic socialism is going to continue to win, and these far-left Democratic candidates are going to dominate: it's the Republicans' fault. They had all 3 branches of government really tightly in their control.

And if you look at what the podcast bros and the pro-podcast-referencing stack are responding to—and I listen to them, not every episode, but I dip in and out of Tucker, who is now making his own political party, Joe Rogan, Tim Dillon, and all these folks who were all in for Trump—they cite the Epstein investigation, which, okay, whatever. I don't think it's particularly material, but it is to them.

The Iran war is obviously at the top of what they believe is the problem here, because they were promised that wouldn't happen. And then it's, to your point, Friedberg, this inflation thing has not been solved, and it just keeps going. We're going to be sitting here, I think, in another 6 or 12 months talking about, “My lord, every investigation possible into the Trump administration.”

We need to start looking at what those things would be, to your point, Chamath. I thought you summarized some of the broken parts of capitalism well, but we have to get granular about it. I brought up last week just the edge behaviors in capitalism where we over-optimize. I felt the delivery providers—and Sacks and I had a pretty hearty debate about it last week—were just being too clever, squeezing too much out of the last mile, quite literally, on Amazon's part.

How we treat the people at the bottom of the employment stack is what's causing this, as are not being able to solve the cost of homes or make that a focus, not being able to solve the cost of child care, and not being able to move the minimum wage. I think we have to look at all of these things and say, “That's…” Or Trump should look at all of those things, and he should build the second half of his term just around those things.

No more crypto stuff, no more wars, no more attacking our allies. All of this performative stuff and these mistakes have to go away, and we have to focus on what he sold people on. People feel they were sold a populist presidency that took care of the working man and woman in this country, and it has been the opposite. Sacks, I'm sure you have some thoughts—

Well, I just want to jump in—

And we'll give you the final word. Yeah.

On the inflation thing, because you talked about it, and Friedberg's talked about it. Basically, you're saying the cost of living is going to drive everyone to socialism. I agree there's some risk of that, but let's break it down into a little bit more nuance here.

If you look at where the most inflation has occurred, it's always in those sectors of the economy where the government is most involved: health care and colleges, where the government has massive amounts of subsidies and regulation, and housing. It only seems to make the problem worse. It makes sense that if the government is paying for something, then the providers have a huge incentive to charge as much as possible. In fact, a lot of them engage in fraud as well, so we know these areas are rife with abuse.

But those are the areas that are most inefficient and where you see the price going through the roof. Hospital services, man, that's just about exponential.

On average.

On average, that's right. But like I said, with certain things like health care becoming almost 300% more expensive, whereas computer software went down by 67% in real dollars. Meanwhile, the average hourly wage grew 135%, so at least the average hourly wage grew by about 40% more than the inflation level. Obviously, it still bites that you have all this inflation.

I guess my point is, how do you solve it? We know that you bring prices down through vigorous competition, not through more government interference in these markets. Unfortunately, what happens is that socialism causes the problem, and then groups like the DSA argue that the way to solve the problem is through more socialism.

So it's a disease that claims to be a cure, and then people take it, and they take more of it, and the disease gets worse and worse. And that is the spiral that I think we risk being in.

We are in a socialist death spiral now.

We're not in a death spiral. Take it easy. We're not in a death spiral. No, I mean, it's great for you to say because it sounds like you got some real hot take. We're not there.

America's doing fine. We are cranking. The economy is great. There are things that need to be fixed. I would say start with corporate America. Let's get our act together, and let's start doing things that are honorable, interesting, and aspirational.

Yeah, I think you totally missed the point. I think the real point here is if he hadn't started the war, we wouldn't have the same inflation issues, we wouldn't have the same expense issues, and this would be a totally different picture. And it was his job, Sacks, when he promised the American people he would focus on affordability. He had control of both houses and the presidency, and he's willing to do any executive order on crypto, on wars, on anything, but he didn't do any executive orders on affordability. And that's the failure of this presidency.

Well, maybe you missed the point where Jamal said it'd be a tailwind to get out of this war. I think the kinetic part's already over, by the way.

Well, let's hope. It just should never have been started.

But in any event, look, the president passed his economic program last year in the Big Beautiful Bill, and I think it's working. We're seeing a huge investment boom. We're seeing small-business confidence rise to its highest level in a long time. They are saying they're going to create more jobs over the next year.

I think the economy's doing pretty well. We do have these issues. I'm not saying we don't, and spending is a long-standing problem. Can't disagree with you there. However, I don't think Democrats have the solution to the problem, and the Democratic Socialists definitely don't have the solution to the problem because they want even more spending.

That's my point about the death spiral—

By the way, have you seen—

—is that we are now incentivizing young people in the Republican Party to embrace socialism. Moderates are embracing socialism. They're all looking at it, saying, “Well, that sounds better. If Trump won't be a populist, maybe these people are the populists.” And that's what I mean by the death spiral, because they're going to just keep spending money. They're going to increase the velocity of spending when AOC is in.

You may be right, which is scary.

I don't think we're in it now. We're risking going into it.

Just one last point. They did a cost estimate of the policy platform proposed by the DSA. Remember, we've talked about that before. And the estimate is that it would project new federal spending—

Hmm.

—of between $71 trillion and $212 trillion—

Under—

—over a 10-year horizon.

How could you spend more if you get rid of all the cops and the prisons?

Underestimate. That's an underestimate.

Wait, did you—

By the way, the total net worth of the entirety of the American population is $182 trillion. $182 trillion. Yeah.

Wait, there's no prison, there's no cops.

Okay, but I'm just saying that's how much the—

Right. There's no money.

—the DSA program would cost. If you guys believe that the root of our problems is overspending and basically too much government, unfortunately—

This is not the medicine.

—the DSA is gonna—

No, but here's what's gonna happen—

—you know, this'll go through the roof.

And let me just walk through the sequence of events. Sequence 1: They enact a policy starting in 2028, with Comrade AOC in charge and the Politburo—Ro Khanna, secretary of Housing and Urban Development, and the rest of the Politburo—organizing themselves to launch these new programs. Those new programs get launched, and costs balloon. The bond market sells off 30-year bonds, and so now the 30-year yield spikes to 7%, 8%, maybe more. Things become more unaffordable.

So then the question is, okay, well, we can't bridge it from a fiscal perspective. We can't just get the Fed to start buying bonds. There's not enough buying power there to bring rates down, to bring inflation down, so we need more income. So now we need to pass a national asset seizure tax, and there's not enough money in the billionaire class or the $50-millionaire class.

The total net worth of all people that are worth over $50 million is only $23 trillion. You could take 10% of their net worth, and it would only cover 3 months of fiscal spending—today's fiscal spending. So at the end of the day, the only way to get there is to start doing massive taxation and seizures.

So this just came out this week—

Cato.

By the Cato Institute, and again, it shows the low end of the DSA spending is $71 trillion. The high end is $212 trillion. Now you compare that to things like corporate profits or private-sector wealth. Corporate profits are only $35 trillion. The Forbes 400, all the wealth of the Forbes 400, is only $6.6 trillion, so it's less than 10% of the low end of DSA spending.

Guys, I keep telling you, this is not about mathematical realism. This is about anger toward a class of people that are increasingly odious and mistrusted.

Yeah, but people also can't afford to buy a house. People are 35 years old with roommates.

And people go to the grocery store, and they're shocked at the fact that they have to spend an entire paycheck or a third of a paycheck to buy groceries.

Also that. It's not that people want to buy a yacht; it's that the yacht class are just despicable.

Well, no, I mean, look, in every socialist revolution, every revolution we've seen historically around this, there is always an elite franch of society, and they're typically classified in different ways. It's wealth, it's Jews, it's some race, it's immigrants. There's some organizational assignment, some identity assignment that says this is the cause of unaffordability.

At the end of the day, the dysfunction of the relationship between government and all the other organizations that make up that society—those organizations can be states, they can be private companies, they can be the NGOs. It is the dysfunction of the government's relationship with the other organizations that fundamentally causes things to become more expensive and dysfunctional, and the actual market for getting access to things for the individual fails.

And at that point, you have to identify—and you don't identify the government, because the government is the aggregate of all the people.

No, of course. That's the Girardian philosophy on this.

So the government—

You have to point the finger, and you have to point the finger at somebody. And if you're going to scapegoat somebody in 2026, the best class of people to scapegoat—

Tech, tech, tech, tech.

—are the billionaires, and frankly, even better than them, the trillionaires.

Yeah, but at the end of the day, it's all tech, right? The data center is the temple where the billionaires and trillionaires pray. It's where they go to make their sacrifices. It's their gathering spot. So I think the data center is this temple on the mount.

It's the avatar. It's a totem.

And so that's why I think the data center—

No, no, no. Look.

—gets tied up in all of this. There's no rational discourse about data centers. It's like—

If you take it—

Yeah.

—through the René Girard lens, everything maps. It's like the billionaires will be scapegoated, and the trillionaires will be, I mean, even worse. And because that will allow the catharsis that will allow society to reset and realize, “Wait a minute. Oh, my gosh. This whole system—”

Yeah. There's a generation of that, though. There's a generation ahead.

It is a long process. It is a long process. Here's an idea: The next president—or Trump in the second half of his term, just putting it out there—could issue an executive order saying that you're going to build, like, 1 million homes, 2 billion homes. If you can have executive orders, if you can start wars without the approval of Congress—

You know that that's not how it works, right? There are local laws.

He pulled out all the stops—

There are state laws.

Of course, but he pulled out all the stops to get data centers green-lit. We know that, and he worked with the state governments. I'm talking about him coming on and saying, “You know what? We did a great job for the data centers. We did a great job for the crypto industry, and now we're going to do 2 great things.”

Well, no, the different and better answer to what you're saying—

Hold on. Let me finish. We're going to do 2 great things. We're going to make trade schools and affordable homes. I'm just going to make that happen in the next 24 months. We're going to figure out a way to mobilize the private sector and mobilize the public sector. We're going to make more homes, and we're going to get you a great education that turns into a job. That would be a great thing for Trump to do.

Well, the federal government does control the largest amount of land in the United States.

Okay.

So it could do something on federal land without the need for a bunch of other state and local gobbledygook. So there is a—

Fannie Mae, Freddie Mac. Yeah.

Now, it may not be located in places that actually make sense for commerce and for people to live, or where people are already living, because I don't think it really abuts most cities. So unfortunately, it's not in his control, even if he wanted to just say, “Let's make this happen.”

I think if the president can go spend a couple hundred billion on this war, he can go spend $100 billion—or a couple billion—on affordable homes.

Oh, so you want the government to build it?

But he could put them elsewhere, and he should fight for the regulations, just like he fought for the regulations on data centers and energy. He made a point on this very podcast at the AI summit: “Hey, we're going to clear all the pathways to nuclear, to natural gas. We're going to have beautiful clean coal.”

They have, right? It's called federalism.

Yes, which is great, and I give him credit for that. Now do it. Now do it for homes. Let me see you do it for homes, Trump. Use that same energy.

Well, they just passed a housing bill.

That's what people want to see.

They just—

When's the last time we heard Trump talk about these things?

They actually just passed a housing bill, just so you know. Unfortunately, it was written by Elizabeth Warren, so I don't know how good it is, but—

Yeah. Anyway, I'm just giving him the solution to save his legacy.

As always, you're looking out for everybody.

I'm looking out for everybody here, folks.

Yeah, you're looking out for everybody. Yeah.

I'm looking out for the audience.

Sure you're looking out for everybody.

Nos Ferkanus. Thank you, Nos Ferkanus. Nos Ferkanus is also looking out for you.

Okay, wrap. I gotta go. Come on.

All right, everybody. It's been another amazing episode.

Love you, boys.

Uh, the ALL-IN Podcast.

Nice to be back.

Good to have the quartet back.

Let's go.

Let's go.

Bye-bye.

Bye-bye.