A.I. Hype Is Running Into Reality

If I put myself in the shoes of, you know, an open-minded, engaged, normy American, and I think, here's this guy who's telling me that, "Hey, I so great it's going to grow at 50% per year, and here's this guy who's telling me, he's going to use it to find a way to fire people." I'm like, "I believe the guy who's telling me, he's going to use it to fire people." I'm David Wallace Wells, I'm a writer for New York Times Opinion, and a columnist for the Times Magazine. I'm Natasha Seren, I'm a contributor to Times Opinion, a lot professor, and an economist at Yale Law School, and the founder of the Budget Lab. We heard today to talk about something very, very big that just happened in the AI economy. Feels like there are many big things, so you're going to have to be slightly more specific. I think we're talking about situational awareness, which is a hedge fund run by a guy named Leopold Ashenbrenner, which made a huge bet on the future of AI. Fantastic name, Leopold Ashenbrenner, 24-year-old, with no actual finance background, who ended up running one of the most significant AI hedge funds in the country and then watched it collapse. The fund had lost roughly $35 billion in assets, plunging from a peak of $45 billion to around $10 billion. So I first became aware of this guy because of an essay, he wrote. The hedge fund grew out of a blog post, which is a remarkable thing given that he also ended up raising tons of money to start the search fund. From lots of it, and we'll hopefully get to it, from lots of names that we know, like Goldman Sachs and JP Morgan. And all these people were reading this blog post, this essay, and thinking the person who wrote this has unique insight into the future of the AI economy, such that we're going to entrust huge amounts of money to his care. So what was in that essay? What did it say about AI? Yeah, situational awareness was actually, seems quite prescient. It was written in 2024, and it kind of predicted that we would be at a moment where, first of all, you'd be by 2027, he thought, very close to AGI, or the idea that we are going to have some sort of super intelligence in these models. He also thought, and kind of understood, before many did, that in order to get from where we were in 2024 to that moment, you were going to need massive capital investments in things like data centers. And that was really going to be imperative to power this boom. And the way the people talk about this is they use the phrase, "CapEx." Correct. And the nature of the hedge funds bet, once he eventually started situational awareness, was about understanding that he was essentially long AI, so making a lot of investments in the types of things that are likely to either profit as we are building out AI CapEx, or ultimately profit as we are deploying this technology. And short, companies like Adobe, where you are worried that the nature of enterprise software is going to be fundamentally disrupted by the fact that artificial intelligence is here. What happened at the hedge fund is actually interesting to try to understand whether it's really dramatic collapse of recent. Is it telling us something about AI, or is it telling us sort of a tale as old as time with respect to how hedge funds like this collapse? Well, my view is sort of that it's both. So, you know, he was incredibly over leveraged. Four times leveraged, right? So, for every dollar that he raised from investors, he borrowed four times that from public markets and from private markets. And that meant that he was really exposed to any even short-term fluctuations in these patterns that he was projecting, which meant that when there were such fluctuations, he was in a really tight spot and ended up having to sell, depending on the reporting almost all or all of his public portfolio in order to cover himself in relatively short order. Also, this happened like three days before his wedding, extra drama, the fact that he's 24 years old. Or letting in Carmel, I think, to the chief of staff at Anthropic. So, it's all kind of this like tangled web of really interesting things. And really incredibly rich people. And so, you know, that's like, as you say, it's a kind of an old Wall Street story, especially when you think that he is this young gun who had come in, you know, was not that long ago being talked about as one of the great success stories of the recent hedge fund world. A thousand percent returns, you know. And it was interesting about it is that in some sense, he might very well end up being right. And what I mean by that is it very well might be true. And in fact, we are watching and have been talking about and will continue to talk about these massive artificial intelligence expenditures, the idea that you're going to start to see productivity gains from automation of certain types of tasks. And that that might very well disrupt legacy software. The problem, and this is again, why I say tails all this time, there's a quote that's attributed to the famous economist John Maynard Keynes that says the market can stay irrational longer than you can stay solvent. And what ultimately happened here is that the same banks that were happy to loan him money on the way up and say that keep making those trades and they're so profitable, that's great. Immediately as it started to look a little shaky, as it started to look like potentially the banks themselves were going to lose money. They made what is called a margin call where they essentially said either you have to give us cash right now in order to protect these positions, or you have to be in a situation where you start to liquidate or sell your assets in order to be able to hand us dollars. And that creates this perpetuating cycle on the way down, right, because if you sell the stuff well, then it actually pushes the price further down such that you have to sell more of it. And that's ultimately what happened. And when Leopold described this, he said it was like a traditional bank run type of dynamic and caused by leverage. We've seen this story before we saw it in long term capital management in the late 90s that was kind of an harboringer of a financial crisis, which is what people are worried about right now. But I think we shouldn't sort of mistake the fact that this hedge fund was over levered and many others might be that are making these trades. What do we know right now about the fundamentals of artificial intelligence, and how is that changed over the course of last few months? Well, the thing that I would say, the reason that I think that it does tell us something about those dynamics, which not to say that, you know, everybody's going to go bust or we're heading for an immediate crash. But the reason that this does raise some serious questions for me is that the story that Leopold was telling in situational awareness matches the story that all of the AI companies have been telling. All of their investors and all of Americans for several years. And in broad strokes, you summarize it, but I just want to give it a compressed version. The story here is AI is completely transformative. It's getting much better, much faster than anyone understands or appreciates. That means that very soon we're going to see a dramatic takeoff and capability. And beyond that point, the economy will be so transformed that the first company is to like cross that finish line, are going to be reaping immense profits. Of a scale like we have never seen before. And when investors hear that, they get excited. When Americans hear that, they may get scared about what it means for their jobs, etc. But it's basically a story of such overwhelming narrative propulsion that all the little considerations, the question of leverage, the question of whether this is going to happen in nine months or 10 months or 12 months or 15 months, all of those things seem kind of secondary. And here we had someone who made an enormous bet. Not just that AI is going to be a big deal, but that it was going to be such a big deal that none of the conventional guardrails were necessary. And that's a big observation. Because two years ago, three years ago, AI boosters were often telling some version of this story. And we're now in a place where I hear many more people and read many more people raising questions about those little things, raising questions about, you know, exactly how much profit has to come into justify the Catholics, raising questions about exactly what it means that they're getting pressured from China. And so on some level at a narrative level, it looks to me like this marks or punctuates a kind of reset where like we're now talking about the AI economy, who's going to win, who's going to navigate that bumpy road and how to allocate resources and capital, how to manage political challenges. And that's a very real world landscape, which is very different from the like whiteboard in a conference room, we're drawing a line on a board and saying that's where we're going to take off. So that's at the narrative level where we've kind of left behind the big story that AI was selling us for several years. And we're now trying to figure out where are we. So Natasha, where are we? So first of all, by the way, the last time we were here, we were talking about SpaceX and its valuation and its public offering. And since that moment, just very recently, SpaceX announced earnings and it announced giant losses on its AI's business, such that the stock came tumbling down. That valuation is somewhere like 50% of where it was when we were first having our conversation. There are very fundamental questions about what AI is going to do to the economy writ large. What is it going to do to our capacity to work? What is it going to do to the labor market? Is it going to displace jobs? Is it going to make firms more productive? That's like one set of issues. There's another set of issues which feels almost both more urgent and more complex to me. Even if you accept that AI is going to be transformational, already has been transformational in lots of ways. If you look at these leading labs and you are getting evidence right now in all different directions, you just heard of recent that open AI and anthropic are hitting these huge revenue targets, even exceeding them that they had set for themselves, which means people are handing them dollars. Anthropic especially. People are handing them dollars in order to get access to cloud code. Okay, so that sounds really good. But on the flip side, you're also hearing about the fact that Chinese open source models, which by the way do not require you to pay those dollars in order to get access to them, are likely to come in and compete away the capacity in the fundamental business model of these leading labs. And even the leading labs are now kind of openly saying they need to compete on price as opposed to quality, which is a sign that this is a huge threat to them. That they understand that it's a huge threat to them and to their fundamental business model. But the valuations that they have where they've been drawing in dollars from investors are kind of have baked into them the idea that they are going to be the winners of this technology and their business models are going to stand. And if that's not true, I just wonder what that means for the economy. In that, imagine, and again, I'm not saying that this is going to be the case. I'm just saying this is one of many plausible scenarios. Imagine you're in a world where open source has competed away their business models. And Anthropic isn't worth that much anymore. And OpenAI isn't worth that much anymore. And if that kind of fundamentally cause a real decline in confidence in the American economy and isn't that going to have the type of systemic consequences that many who are wondering, are we in an AI bubble, are we not? You actually start to get those even if AI in general is going to produce massive productivity gains and massive profits, but if those labs fail, what happens feels like a really fundamental question. And I think many Americans would also have big questions about how we found ourselves in a political economy that allocated so much capital to- So quickly. Disprojects. You know, relatedly, and we're talking a little bit about leverage, you have these large technology companies that are called hyperscalers that are really building up the data center infrastructure that these labs, agents, and models are deploying. You have them for the first time for years. They have been like sitting on piles of cash. Now you're in a situation where they are taking on tons of debt in order to finance exactly these investments. And I think- And there's some debt even that's off the books, right, that we don't say, you know? In these special purpose vehicles that again, like, starts to harken back to like the financial crisis. It's not a sign of health. We were- We don't have huge amounts of off-book debt, right? So I'm going to make the arguments both ways in that like off-book debt sounds bad, harboring or financial crises. Flipside debt only loses and loses value. Once equity, so the investors who have handed dollars, are wiped out. So you have to think that there is like a fundamental threat to the business model of like Google in order to be super concerned about this leverage building up and I'm like less convinced about that. So let's think about this a little systematically, right? We're talking about the risk that AI is overvalued in a sort of systemic way. Maybe that could lead to something like a bubble popping, maybe just like a lesser correction, but some turbulence ahead. And when we think about that risk, you know, I see the revenue for Anthropic into some lesser extent open AI, arguing that things are pretty good actually, that there's like this is like, you know, a good gravy train to be on as an economy as a whole. And then on the other side, there's a lot of stuff happening that suggests some concern. And I wonder if you could walk us through those worries. When you're thinking about the risk that we're heading towards some adjustment, negative adjustment, what are the things that you're focused on? What do you point to as signs of concern? So the first we've kind of started to touch on already, which is that it is true. That open AI and Anthropic especially, they had like a banger, July's, they way exceeded their revenue expectations even for themselves, which were quite high. And so they feel like they're answering the question, how are we going to generate profits that justify these sky high valuations by like saying, look at the data. And we in fact are generating those profits and then some, but I think this question about how competitive this industry ends up being and already is frankly with respect to not just Chinese open source models, but open source models writ large in the idea that actually like you're going to be quite content with a slightly less good model that you can get and then monetize if you're a firm and deploy in a much cheaper way is ultimately going to win the day. But getting from here, where these labs and these valuations and these dollars have been invested in such a dramatic way to there where they are not the main players is going to cause some market disruption. I think the second big risk is if you look at how profits are being allocated in the economy at the moment, the people who are doing like super, super well are like Nvidia. They are the ones making the infrastructure that powers this a I boom, they're making the chips, the ones who are making the data centers, that's like great. They are actually doing better than the types of firms that are the employers of this technology. And by the way, a lot more of it is being debt financed now than it was before. And so that raises another concern, which is if these bets don't quite play out, if it turns out the data centers take longer to get online, if it turns out that actually we've overbuilt and we have too much capacity, if it turns out that there are political and regulatory barriers that we haven't quite yet imagined. In those situations, you're going to start to be in a dynamic where you very well might have dollars that have been borrowed that can't quite be paid back. And that starts to get concerns about like some sort of systemic crisis in the economy that spreads more generally. It starts to invoke those types of concerns. And by the way, if you're not a subscriber, we have some news for you, you can now explore the New York Times for free without paywalls during your first month in the New York Times app. I wanted to drill down on two particular points. The first is about the arrival of open source threats to the Frontier labs. And that I think is significant on a bunch of different fronts, some of which we've talked about, but one of them is to me, the political economy part of it. Now if I think about why a lot of people in Silicon Valley are signing on to an open letter about to say we shouldn't fight open source, we shouldn't fight Chinese models, we should let them in. And some of that is naturally just the competitive instinct of companies that are not anthropic, which has a closed model and is doing extremely well. But at the level of like ideology and shaping, you know, the medium term future of the economy, if it is the case or becomes the case that the Trump administration overall, the AI industry as a sector, if those, if we see much more openness to this tech, where does that lead us? God, there's so much in that question, David. So first. I mean, we've been talking for years about this as an existential cold war level threat or race with China. Now we're like, maybe we should just let them in. Actually, China sounds good. Yeah. So there are many parts of your question that like feel super important to me. One is we haven't really talked much about the nature of the security threats that AI potentially poses. And those have different flavors, one with respect to Chinese open source models. And there's another type of security threat that's like also super important. And you're also starting to see like real sort of fractures in which is we don't quite have like full control over what these models are actually doing where you're watching these models like fundamentally break out of enclosures that have been set for them by humans. And do things and operate in ways that they've been explicitly told not to do. And, you know, the real concerns about like, are these models building the capacity to build biochemical weapons, right? And Bob-- - We used to talk about those risks so much. - So much. - The existential risks, the bio terrorists. - At the forefront of the conversation. And they should, in fact, be at the forefront of the conversation. And part of what I find so interesting is that, and maybe this is somewhat helpful about Chinese open source in that, if you talk to and hear from a lot of technologists in Silicon Valley, you and I talk to a lot of them, they will tell you that there needs to be some, because of these risks, some sort of globally coordinated regulatory framework. - Well, sometimes they'll say that. - Sometimes they'll say don't touch us. - And some of them will say that, but like, it's always struck me as kind of like nuts, because we know how the political system works in the United States, which feels kind of dysfunctional. The idea that we're gonna be able to develop the political capital to do like a whole global conversation about AI and what sort of structures we should put around it hasn't really seemed that likely to me. But if you're in a situation where the United States in China and actually give the Trump administration credit here, they've at least announced ostensibly a desire to have exactly these conversations about what it looks like to actually try and create some sort of agreement or structures about how the technology should ultimately be deployed and what guardrails should be put around. - It's actually been a really interesting story of this 18 months of the Trump administration. They came in, they seemed like they were technological accelerators. They wanted to rip off all the, and they've made a pretty serious evolution, even just in this year and a half that they've been in power. And are now, they're not like, you know, on the safest end of the AI safety spectrum, but they, but they're grappling with these questions, you know? And I actually do think that there is some capacity. I mean, in some sense, we all, we are all bringing to this our own interests, but like China's an authoritarian government who must also be concerned about the idea of technology developing capacities that it's not able to control. And so in some sense, theories, hopeful story is that there is some sort of motivation to come to the table in a way to deal with exactly these types of real existential concerns. And open source might actually give you an entry point and a necessary entry point into those discussions. - Yeah, I mean, one aspect of the China US contrast that's always been interesting to me is that, you know, the US is sending much more money here than any other country in the world in the build out, but we're also a country that's really anxious about AI in China, you know, they're kind of in second place. The country's much less anxious about it. - For, you know, and there are a lot of things going into that, but one of them I've always thought is that they trust that their government is capable of taking control of their economy. And here in America, we basically, but that raises the last thing I wanna ask you before we move on to the bleak part of the conversation. - Yeah, so good, we're getting bleaker. - The last part I wanted to, the last question I wanted to ask you about the now, is about the political situation in the US. So this has been something that I've been following, writing about now for a while, and it still astonishes me almost every month. There's a new poll showing, you know, the incredible resistance to data centers. - AI is less popular than ICE, you know, versions of that. And the booing of the grad speakers and all of this. - I mean, I would go even further. We now have like real policy implications here. In New York, there's a moratorium on data centers. And they just this week announced that in Texas, they're putting a pause on attaching data centers, new data centers to the grid. Which is especially striking to me, because when I first started thinking about, reporting on writing about data center backlash, I would have said to you, I maybe did say to you, and I don't remember, I would have said to you, one of the things that's driving this, is that people are seeing an oligarchy building a new economy over which they have no democratic control. And they are expressing that anger in these town halls. But at core, it's about the fact that the future is being built without their consent. And what we've watched over the last few months is actually an incredibly, on some level, inspiring populist backlash. I have, you know, questions about exactly what the agenda there is. I think people are confused about some of the environmental issues. But nevertheless, we have seen a significant uprising against this technology and the economic future that it promises, despite the fact that not that long ago, I think many of us, maybe even you and I, were looking at it being like, I don't know if anybody's gonna be able to take control of this. And as a result, we now have in state after state and community after community, really meaningful obstacles being put up to the build out. Now, I don't know how that looks heading into the midterms. I don't know how it looks heading out of the midterms. It seems to me like, you know, one of the key stories in Abdul Al-Sahid's success is resistance to data centers. Certainly it seems to be a big part of Francesca Hongs, sort of apparent success in Wisconsin. You know, we're seeing politics being shaped by public resistance to data centers. And that is a big question mark going forward. But if they are actually able to take command of the levers of power sufficient to block that development, you know, that is a big, big challenge for the subject of this conversation, which is, you know, the medium term future of the AI sector. How do you see that? - I'm gonna say two things, one about what this means for the labs and the other about what this means for the political project of AI in the United States. And why I am a little bit more worried than you about the data center moratorium enthusiasm. If you go back to this conversation, we're just having about China and the US and the nature of the importance of winning this AI race, which has categorized a lot of the conversations that people have been having about AI over the course of the last many years. The reason why it is actually fundamentally good for the US economy, that we are at the leading edge of this technological revolution, that the investments, the development of the models that are driving AI's growth are happening here, that the dollars are being deployed are being deployed here. These are like fundamentally good things. - And a lot of the industrial capacity is being actually built here. And they're even people who say like the models don't even really matter what really matters is just how much compute we're building. - And by the way, we might actually, into some of the environmental things, as we build out the sort of power grid to sort of support those investments, you actually might see in the medium to long-term, you're actually decreasing electricity cost because you're building out capacity. And so all of that is like good economic story. I have been struck by the fact that if you try to understand why is it the case that you have young people booing down commencement speakers who bring up AI, or some of these polls that show that AI is less popular than pick your unpopular thing. I actually think there is a fundamental communications problem that exists among many of the leaders of these institutions and of this technological change in that they have been saying things like AI is gonna displace 50% of the work that you college graduates are doing. And so who is gonna be in favor of a thing? If instead they had been saying AI is gonna deliver all of this growth, it's gonna deliver all of these profits. It's gonna fundamentally change the way that we think and the way in for the better. Just like the internet changed the way that we live for the better. That could be a message that people could buy into. It's just not the message that's been delivered. - Well, I think some of them are saying that, I mean, I saw, you know, Dennis Asabis who just stepped down from Google as part of this turmoil has said he thinks that the AI revolution is going to be 10 times as significant as the industrial revolution at 10 times the speed, which somebody ran the numbers and it's like this suggests 50% GDP growth here on your-- - Yep, it is, which is nuts. - Well, that's what I mean. And so when they tell you that you're like, that's ridiculous, and when they tell you, we're gonna use this technology to find ways to downsize the workforce, you're like, that sounds credible coming from billionaires. Now, it's not to say that exactly what they're saying that we're gonna lose 50% of what I call it work is all that plausible. I'm personally skeptical, but if I put myself in the shoes of, you know, an open-minded, engaged, normy-american, and I think here's this guy who's telling me that AI is so great, it's gonna grow at 50% per year. And here's this guy who's telling me, he's gonna use it to find a way to fire people. I'm like, I believe the guy who's telling me, he's gonna use it to fire people. - The thing, so I agree with that. And I actually, but I guess what I'm trying to say is that it strikes me as like pretty, it's both pretty important from the perspective of like, how does AI realize the profits that are baked into these valuations that we started talking about? And to like, when does the market correct? The political economy question is super important. Because if this resistance builds such that, and you've seen versions of not just the data center moratoriums, you've seen politicians propose things like essentially trying to put a stop to the technology, whatever it is that that means. If that enthusiasm builds, then we are not going to be able to to reap the economic gains that AI promises, because this isn't a closed economy. This isn't like the rest of the world isn't gonna move or China isn't gonna move. It is that we are gonna slow and no longer be at the forefront of all of the benefits that the technology can help us realize. And what I think the fundamental challenges for the political project right now is articulating those benefits in a way that feels tangible to people like the potential job loss feels tangible to them. But also ensuring that we have a government in place that is capable of building guardrails that's capable of actually providing unemployment insurance and worker training. And like no one trusts that we have that, which is why we are in this situation of, I speculate that is why we are in a situation of seeing this much resistance to something that fundamentally like we should be cheering as like this great technological progress. - Yeah, I mean, I think it's the reason that I'm skeptical that resistance will decline. I mean, I could be wrong. I don't have a crystal ball. I'm gonna ask you in a minute to look into your crystal ball, but I don't have a crystal ball. But what I think is seems likely is that this becomes a bigger obstacle imposing more costs and making the build out of AI infrastructure slower, which at a local level will probably make some people happy, but at a systemic level is gonna introduce an additional layer of risk. - And then there's like a cyclical thing that could happen, right? Like kind of like the circular financing where that resistance makes the, somehow impedes the build out such that you're then in a situation where some of these investments start to sour such that you're then in a situation where the market starts to correct because exactly of those risks, but then that amplifies 'cause then it's like the technology which we don't even like is also leading to this negative economic impact that we definitely don't like and potentially a downturn or a recession. And so it all has this like kind of like self-amplifying effect that feels important. But one of the things you said, David, is like that I will be paying a lot of attention to is what shape does this type of political push actually ultimately take? In that how real are these moratoriums? And what type of sort of pathways are there to work around them? And ultimately, what capacity do we really have to constrain technology and technological progress even if many would want to? Feels like an important question that I just don't know the answer to right now. - Okay, so we'll pull back from these unknowable questions about politics and focus on the incredibly noble stuff about how the market is going to evolve and how we can all make a killing going forward. So if we're thinking about the possibility of a market correction here, like some meaningful change in the basic fate and valuation of these companies and the sector as a whole, like what should we be looking for? How will that play out? If we're like imagining a scenario in which the problems that we've identified are serious and are really getting in the way of the economic promise of AI in the medium term, where are we going to see that show up and how all it shake out for the rest of us? - So I think you should be looking at a couple of things. I think one of them is kind of boring in that you should be looking at like our solutions to a basic physical problem. Are we able to build power plants and plug in these computers at the speed that has been promised? And like again, we'll be able to in live time see the extent to which that is true. And once you start seeing some of those targets miss and some of those delays occur, you should be a little nervous. I think the other thing is, and this is why I sort of started by talking about open AI and anthropic and July, looking very good from a revenue perspective. I think you have to start to look at the question of what types of revenue gains are you seeing and what types of revenue gains are you not seeing? That leads you to some concern about the nature of whether there in fact is mispricing or overhyped asset balance. - I mean, it's telling that you're even saying revenue is not profits, right? - Totally, absolutely. And I think that we are in a situation where for me, one of the bigger questions has been continues to be, but is with more urgency now than we last spoke. Like if the fundamental business model of a lot of these biggest labs kind of works. And if it doesn't, so in some sense, I speculate, I guess where I'm blind my biases as I talk to you in that I actually, I think that we are very likely over the medium term, long term, whatever you want to call it, to see pretty significant economic growth on the heels of this technology. I think it is less clear who the ultimate winners and losers in the economy are likely to be as a result of that growth, and whether at the end of all this, some of the names that we talk about all the time are gonna be the most significant players, particularly in a world where open source feels as important as it does, and that strikes me as a place that will ultimately cause the market to correct because they're so important and so significant at the moment, and if they start to look less significant, those valuations just cannot be justified. - Yeah, I mean, it's another way in which the story that I was telling at the top of our conversation about the narrative reset seems to hold. Like it seems possible that in a relatively near future andthropic and open AI are not gonna be essential to the way that we're thinking about these things as has been the case for the last couple of years. But, you know, it also makes me wonder if that, if we're expecting something like that to play out, where AI is here to stay, it's significant, it is meaningfully driving economic growth, but we see the decline or even collapse of a couple of these massively valued companies. What does that look like to the average American? What are the ripple effects to people who own stock, but also just like people who are looking out at the unemployment rate and GDP growth like? - Yeah, so like a couple of things. One is, I wish I had like a satisfactory answer for you in that these are exactly the effects, but fundamentally we have no idea. And in part, we have no idea because of some of the things that we've been talking about, about where do the dollars sit that are financing these investments, as these firms go south, who loses, who wins, we don't really know the answer to those questions. But something that should give you a little bit of comfort is that in some sense, if you think about what happens when a firm fails, the people who stand to lose the most are the shareholders of that firm who are gonna watch its value deplete. And so in some sense, those are relatively, that's like we're worried about Elon Musk losing a lot as SpaceX valuation is decline. It's true that we all have exposure to these companies in our portfolios because of the nature of the fact that especially as they go public, they're listed on these indices, but even as they're private by the way, our pension dollars, our insurance dollars are invested in them. But whether that actually translates into the kind of systemic financial collapse that you get when you've heard of some of these bubbles popping in the past or things like the great recession, I think there are a lot of reasons to be a little less concerned that that is likely to play out this time around. Just because of the fact that we've talked about, which is that a lot of the investments are being driven by large technology companies that just feel fundamentally different and feel fundamentally on some level more likely to be able to sustain their business models than maybe some of the labs whose work they're ultimately funding. - It's like the self-dealing of these companies is actually a safeguard against it. - Yeah, on some level. - Onishing us. There's also I think a kind of a possible cultural fallout which is to say, if there's a dramatic change in the AI outlook, even if it doesn't produce a huge market correction at the level of 2008 or 1999 or whatever, people will still think, what was that? Like those five years when all of our cultural capital, all of our literal capital is being dumped in a bucket. - And these people and these characters that are driving these stories, it's so weird, right? - I think it could produce a kind of significant cultural backlash not unlike what happened after 2008, even if the consequences aren't that bad because people will just say, why were we told that this was the future? Why did so many people bet on it when we could have been investing in other ways? Now, like you, I have complicated feelings about this whole landscape and where it's heading. And I think there are probably meaningful benefits that are coming our way from this technology. I don't want to sound like a Luddite on it or even a populist. But I do taking the sort of sense of the way the wind is blowing. It feels likely that even in a relatively good draw here, we still may be producing some meaningful public resentment and background. And you already are. And in fact, you're sort of view that like should we have been directing all these dollars in these places? I mean, that's very consistent with what happened in the dot com bubble, right? Where exactly it was that we under invested in certain industries and sectors because like dollars were flowing into pets dot com. And so I think those are like, it's not just that people will feel that way maybe. It might very well be that those are like real and legitimate concerns about the allocation of resources in our economy. And it's part of what like really worries me about this moment is I think it's going to ultimately be very hard to run the counterfactual and say like, what would this time of look like if we like thought about the world slightly differently or, yeah, or these characters in Silicon Valley weren't commanding so much of our attention and so many of the podcasts that we're doing. But it's just like, it seems striking to me and it seems like a pretty weird time. Well, I think that's a good place to leave it. So Natasha Seran. Thanks so much for talking. It's been great. Thanks so much for having me. [MUSIC]