The Semiconductor Century with Chris Miller (Author of “Chip War”)

Speaker A: In the future, China is going to try to use every possible way to break through the barriers that the US has set up. And so my guess would be that in 5 years, China is still behind TSMC, but there is uncertainty there. And if China were to catch up with TSMC, it would have really profound implications. Speaker B: To catch up to US capabilities and TSMC's capabilities, is your expectation that China has to essentially create an alternative to ASML, like EUV? UV machine? Is it that they somehow managed to get one into the country in some way, shape, or form?

Speaker A: There's no way that smuggling is going to solve the problem. These machines are far too big to smuggle. There aren't any sitting around unused in the world. And so the question is, can China either A, replicate everything ASML has done, or B, find an alternative way of producing very small transistors on its chips? Speaker C: Hey, I'm Mario, and this is The Generalist Podcast. You've probably heard the saying, the future is already here, it's just not evenly distributed yet. This podcast is my attempt to map the future by interviewing the people creating it, driven founders, prescient investors, and brilliant thinkers.

I hope my conversations with them help you identify significant changes before they go mainstream, make sense of complex developments, and find ways to participate in the upside. Today I'm speaking with Chris Miller, the historian who wrote one of the most important geopolitical books of the past few years, Chip War. It also happens to be one of my favorites. Chris is a professor at the Fletcher School at Tufts University who has spent nearly a decade showing us how semiconductors, the tiny chips powering our lives, have become the strategic center of gravity for global power economics, and innovation.

In our conversation, we discuss the China-US semiconductor race, where Beijing is achieving self-sufficiency and where it still lags, what most people got wrong about DeepSeq, but why it still matters, Huawei's surprising transformation from a telecom giant into a potential AI powerhouse, and why India and Southeast Asia might be critical players in the geopolitical chess match of chip manufacturing. Speaking with Chris reframed how I think the next decade may play out as it relates to computing and great power conflict, and I hope it proves similarly valuable for you. This is a new podcast, so if you like it, I hope you'll consider subscribing and joining us for some of the incredible episodes we have coming up.

Now, Here's my conversation with Chris Miller. This episode of The Generalist Podcast is brought to you by our very own Generalist Plus, the premium newsletter that's redefining how investors and builders navigate the technological frontier. Generalist Plus delivers a mini MBA to your inbox at just a teeny fraction of the cost, just $22 a month or $220 annually. So What's included? 1, tactical interviews where elite founders and investors reveal their actual strategies and decision frameworks. 2, comprehensive guides that distill hundreds of hours of research into actionable insights on investing and company building.

3, an exclusive database of emerging startups poised for significant growth. And finally, complete access to our archive of meticulously crafted case studies. All of this comes wrapped in the distinctive storytelling and incisive analysis that readers have come to expect from The Generalist. We've designed Generalist+ to level up your capabilities as an investor and operator through knowledge that matters, delivered with precision and depth. So join a community of strategic thinkers who are gaining an edge in understanding markets, technology, and business fundamentals. By visiting com. That's com. Speaker B: Chris, thank you so much for being with us today.

Speaker A: Thank you for the invitation. Speaker B: I think many of our listeners will have read Chip War already, or at least I, I certainly hope so. But for those who haven't, perhaps you could give a brief distillation of what I think is perhaps my, my favorite modern history book. Speaker A: Well, first off, thank you for that. Um, I, I decided to write Chip War now almost a decade ago when I came to realize that you really couldn't understand how the world worked without semiconductors, whether it was the shape of the global economy and the supply chains that crisscross the Pacific Ocean, or the balance of military power and the use of, uh, capabilities like computing and guidance to transform how militaries fight, or the future of artificial intelligence.

All those, uh, would already have been and, and will in the future be defined by access to computing power. So that's why I wrote Chip War, to argue that even if you weren't in computing or in semiconductors or in software engineering or anything related to computing, you in fact were living in a world that was profoundly shaped by many, many billions of chips. Speaker B: And one of the things that I love about that book is you really captured so many of the human stories behind it. It is a world awash in many rich characters, and we're seeing sort of new chapters of that, that story play out today.

When you started it, you know, working on it 10 years ago, did you have any idea that it could be quite so important to understand this world? It feels like the last, you know, 4 or 5 years in particular have made this essential reading, essential to understand. Speaker A: You know, I think I had a dim understanding of the direction of travel of some key trends of the proliferation of chips throughout supply chains of, the growing importance of large volumes of computing and artificial intelligence, but I certainly wouldn't have predicted the way all of these trends converged just over the last couple of years to put chips at the center of public discussion, of financial markets, of how we think about the future of technology.

So it's been an exciting couple of years to watch all that play out just so rapidly. Speaker B: Amazing. Well, let's talk about the future. I was having a conversation with an investor the other day, an investor who's very involved politically also, and, you know, is charting the changes in administration, has a, you know, a broad macro view. And his take was that 2025 is really the mark of a new era in history. As a historian, I'm curious what you think of that. Are we, are we in a new era?

And how does How does that impact the chip war? Speaker A: Well, I think one of the things historians love to do is to define new eras, and you can find historians who will describe almost any given year as the start of a new era, at least from a certain perspective. I, I think when it comes to, to, to politics, I think it's easy to always overestimate the novelty of the moment that you're in. And so I, I would just have an intrinsic skepticism that in a political sense, we're in a fundamentally new era.

Although certainly a lot, uh, uh, will change with the new administration. I think there's more of an argument to be made that we're on the cusp of a new era when it comes to economics and technology, depending on, uh, your projections for just how impactful AI will or won't be, both in a broader macroeconomic sense as well as in a, a technological sense. It'll clearly be impactful. Um, but will it be a, a continuation of a trend of applying computing to more and more segments of human life? Or will it be a major step change upwards that creates really significant short-run, um, acute, uh, social, economic, political dislocations to the positive and the negative direction?

I think it's, it's still unclear, to be honest, but I could certainly see an argument that we're at the cusp of a new era driven not by the politics, but more by the technological change and its economic implications. Speaker B: What are the sort of signals you're keeping an eye out to decide which of those two paths we look set to head onto. Speaker A: You know, I think one of the takeaways that I've drawn from my historical research is that far too often people over-index on technological change as a proxy for the impact of technology on a society.

And instead, what you want to index on is the diffusion of technology, because ultimately that's how technology has these broad social, political, economic impacts. And so when I look at artificial intelligence today, for example, you know, you can try to assess how are models improving on benchmarks or, uh, try to find, um, ways of measuring quality, uh, today relative to several years ago. And I bet those are the wrong metrics to be looking at. And instead, I think the question is, uh, how rapidly are new capabilities being diffused and adopted, uh, throughout society?

Because ultimately that that's what gives you the, the social and the economic impact. I think the challenge is what are the specific numbers you would be counting, uh, to measure diffusion? That's a hard question to, to actually answer. You can look at number of subscribers to ChatGPT, but that's probably a very, very loose, uh, proxy for what you're actually trying to get at. But it, it does seem to me that the question of diffusion, uh, and of adoption is far more important, uh, than measures of capability per se. Speaker B: Yeah, ChatGPT subscribers, the Claude users, interesting proxies, more even just taking a look at the extent to which big businesses seem to be happily adopting all manner of different products.

It feels like there is, you know, insane pull from these large institutions to figure out how to trim a customer support team from 1,000 people to 10 or you know, integrated into different parts of their product. Are those some of the other pieces of this that you look at to sort of gauge diffusion? Speaker B: Yeah, ChatGPT subscribers, the Claude users, interesting proxies, more even just taking a look at the extent to which big businesses seem to be happily adopting all manner of different products. It feels like there is, you know, insane pull from these large institutions to figure out how to trim a customer support team from 1,000 people to 10 or you know, integrated into different parts of their product.

Are those some of the other pieces of this that you look at to sort of gauge diffusion? Speaker A: Yeah, I think that's, that's, that's right. And the intuition there being that, uh, there will certainly be consumer products, uh, leveraging AI already are. Um, but perhaps it's the case that the, the primary impact, at least in the short run, will be in the enterprise. Um, and I think that's, that's an intuition that makes sense to anyone who's in the business of selling software where the enterprise market has been a very attractive market for a long time.

but actually isn't how we publicly talk about, uh, the impact of tech diffusion. Um, and, and, and that I think speaks to a different set of capabilities that companies are looking for versus what individuals are looking for. And I think even within the enterprise, it's difficult exactly to know what metrics are the most impactful. Um, but I, I certainly do find myself, uh, thinking very hard about what are those metrics? What are the anecdotes in the business, uh, um, segment that I think are the most impactful? And, and index on those rather than the consumer dynamics to think about how rapidly is diffusion happening or not happening.

Speaker B: On a personal level, I'm curious how much you find, uh, AI has diffused your time. Like, you are a, you know, an information-consuming machine and a synthesizing machine. On our last time we chatted, you were talking to me about, you know, how you love to go down these rabbit holes of, uh, Japanese company histories that haven't been written and so on and so forth. Yeah. Versus Chris Miller 5 years ago, where has it sort of seeped into your, uh, day-to-day most? Speaker A: You know, I, I, I do find myself using ChatGPT in particular as a, a complement to Google, Google Search on any sort of, um, anything above the simplest of questions.

I usually ask Google and ask ChatGPT and see how the results are similar or different. I, I recognize that's a very first order, um, use case, uh, but that's where I've found the easiest integration into my own workflows. And then I, I think I'm, I'm eagerly awaiting, uh, new products that, uh, seamlessly integrate for me more agentic capabilities. For example, they haven't really arrived yet. I think, uh, I think we're, we're still in, in waiting mode. I can imagine many ways that I will be an eager user, but I think the productization actually matters a, a lot there.

And, and so right now I'm, I'm primarily a, a user of AI systems to just give me a different cut of information than Google would give me. Speaker B: Talking about tech diffusion feels like a natural segue into Deepseek. Um, and I'm sure you've talked about this and thought about this a great deal. Uh, for those who maybe didn't follow this story, perhaps you could give us a sort of brief précis, and, and more importantly, I'd be curious for your take on how significant, uh, the, the announcements that came out earlier this year really are.

Speaker B: Talking about tech diffusion feels like a natural segue into Deepseek. Um, and I'm sure you've talked about this and thought about this a great deal. Uh, for those who maybe didn't follow this story, perhaps you could give us a sort of brief précis, and, and more importantly, I'd be curious for your take on how significant, uh, the, the announcements that came out earlier this year really are. Speaker A: You know, I think Deepseek is certainly significant. It's, uh, an AI lab in China spun out of a quantitative hedge fund.

Which until about 6 or 8 months ago was really not anywhere close to the center of anyone's mental map of AI in China or even globally. Um, but over the last 6 months has been releasing a series of, uh, pretty impressive models. And then in January of this year, rocketed to the top of global public attention thanks to the release of its latest reasoning model, uh, causing in one day a 17% decline in NVIDIA stock price and, uh, many other implications downstream of that. I, I think Deepseek is, is certainly meaningful in, uh, in two ways.

One, it, it shows, not that this is really a surprise, but I think it's a useful reminder, uh, to many people that China's got extraordinary AI talent. And two, that when it comes to algorithmic design, the moats around algorithmic design are not very deep, if they exist at all. And this is because, uh, A, if you've found out what your competitor's algorithm looks like, you can copy it straightforwardly. And B, the research that's involved in finding the next improvement to the algorithm is itself not very heavily defended by, by unique moats.

It's most of this research is still published on the internet, not all, but most. And there's many different avenues for improving AI systems. And so as a result, um, anyone who is hoping that their unique algorithm or their ability to develop ever better algorithms would, uh, be a, a strong competitive, um, uh, mode against, uh, against potential rivals, I think was, was proven wrong by many things, but including by, by Deepseek. I think there was an alternative narrative though that Deepseek had managed to train a cutting-edge AI model with a tiny fraction of the computing power, the chips that a company like OpenAI or Anthropic had.

I think that's basically not true from what we've learned, is that actually they had access to a fair amount of compute. They defined the amount of compute they used in the most narrow way possible to exclude all the other R&D they did. And so I, I think the takeaway that led everyone to sell their NVIDIA stock was probably wrong, but the takeaway that actually there aren't good barriers around algorithms and that one should expect rapid diffusion of algorithmic efficiency improvements, that I think is the right takeaway. Speaker B: So many threads I want to follow there.

On the, you know, sort of overreaction to this. 5, 6 million number that was going around saying, you know, Deepseek achieved this with, with this tiny amount of money. Do you think that was intentional in any way to lead with that number? You know, there was certainly some chatter around, you know, is this Chinese psyops? Is this, you know, um, some kind of gameplay, uh, in the works? Like, how do you make sense of that having studied so much of China's work in the semiconductor realm? Speaker A: You know, I certainly noticed the fact that the Deepseek model that generated so many attention, so much attention was released the week of President Trump's inauguration, as well as the week of the Stargate announcement.

Was that a coincidence? I don't know. I haven't seen definitive evidence on either side. And so I'm going to reserve judgment. The alternative explanation is that it was released right before the Chinese New Year when everyone went on holiday, which also seems like a plausible explanation. So I'm, I'm, I'm in a waiting mode for more evidence before I have a strong view on that issue. I think Deepseek is, is interesting in the Chinese context because until pretty recently, from all the public information we have, when the Chinese government thought about its leading AI players, Deepseek was not on the list.

And it's only in 2025 that Deepseek leadership has started getting meetings with senior Chinese officials, and now it's had a couple, um, already. And so I, I think it's the case, and again, if we have alternative evidence, I'm willing to adjust my hypothesis, but from what we know right now, the Chinese government wasn't really thinking about Deepseek much until at the earliest late last year and possibly even early 2025, which, which again would make me think that there, I think the base case assumption is that there wasn't a whole lot of coordination, even though I'm open to the possibility that that there was if we see new evidence.

Speaker B: When you talk about the sort of major players within the Chinese AI community historically, who, who would have been on that map? Who, who are you sort of thinking of? You know, I know that in, in the book you certainly talk about the efforts of SMIC and, uh, you know, uh, you'll, you'll have to remind me of the name of the sort of memory player in the space, but, uh, yeah, curious what that landscape has traditionally looked like. Speaker A: I think there are the chip players, SMIC, um, The producer of, uh, high bandwidth memory in China, uh, CXMT.

Um, but in addition to the chip players, there are also the, the big tech companies. So, um, Alibaba, Tencent, Baidu, all have spent a lot of money, hired a lot of, uh, very smart people to work on, uh, AI. There's ByteDance, which, uh, isn't a publicly traded company, so we know less about their spending, uh, in this sphere, but by all accounts, they spend a lot of money, maybe even more than the others on AI infrastructure, building data centers. Then there's Huawei, um, which a couple of years ago would never have described itself as an AI company, but now is really rebranded, um, and, um, both publicly describes itself, and I think is right to be described as, uh, aspiring to be a full-stack AI company from the hardware all the way up.

Uh, and, and those were the big players. Those are the companies that were profiled in the Chinese media. Those are the companies whose CEOs were meeting with Chinese leaders. And Deepseek was really not on that list until the last couple of months. Speaker B: You know, you talk a lot about Huawei in the book and what a sort of phenomenal story it is. Uh, I haven't really tracked that transformation. Is that a transformation that feels credible to you, that they can play as this, this sort of full-stack player? Speaker B: You know, you talk a lot about Huawei in the book and what a sort of phenomenal story it is.

Uh, I haven't really tracked that transformation. Is that a transformation that feels credible to you, that they can play as this, this sort of full-stack player? Speaker A: I think it's at least, uh, it's at least partially credible. Certainly at the, the chip level, um, they've got the most advanced, um, chip design capabilities in all of China and capabilities that are, are globally competitive, maybe not at the top, but globally competitive. Um, they've been building out their chip manufacturing supply chain very rapidly over the last couple of years with a handful of different partner firms across China.

They realize that one of NVIDIA's key assets is CUDA. And so they've been putting a lot of money into building out, um, all of the software ecosystem around their own GPUs. It's not nearly as good, um, from all the information we have, but nevertheless, they're making a real effort. At it. I think as you get further up the stack towards more, um, um, uh, consumer-level applications in particular, but also enterprise software, Huawei is a much smaller player. Um, but I also think those are parts of the stack where if you, if you dominate the bottom and you've already got some good distribution channels on your phones, for example, uh, I think it's not necessarily that hard to make yourself a, a significant player in the provision of simple AI models for smartphones if you already dominate the smartphone market.

in China. And so I wouldn't bet against them in terms of their core technology capabilities in general. And I certainly wouldn't bet against them in terms of their Chinese market share, where they've got a very strong competitive position. I think the question is, can they export in AI-adjacent products? Can they export GPUs? Can they export other AI products in the way they succeeded in exporting lower-level sophistication electronics like telecoms, uh, systems like smartphones. And, and that's where right now we haven't seen any evidence. Um, but we'll— I'll be watching closely the next couple years.

Speaker B: We, we talked, uh, we, we started our discussion about Deepseek talking about, you know, tech diffusion. I was reading recently, I think it was in Marginal Revolution, Tyler Cowen talking about, you know, Deepseek's real impact will be, uh, in Nigeria, in, you know, large African markets. That, you know, have maybe been cost constrained in, in accessing the more expensive models. Is that sort of a view that you would go along with? Does that make sense to you that, you know, there is this opportunity, uh, for, for Deepseek and, and other lower cost models to win some of those emerging market use cases?

Speaker A: I'm not sure that I would, I would have that as my base case. I think it's true that Deepseek offered low prices. It's also true that since, uh, 2023, we've seen an absolute collapse in model prices. First for GPT-3 quality models, then for GPT-4 quality models, now in reasoning models. And so, does Deepseek have low prices? Yes. Uh, is that part of a broader trend that many companies have participated in? Yes. And so, will Deepseek have the lowest prices in 2026 or 2027? That's far from clear to me.

Uh, and I think I would have the view that unless we have a rapid decelerate, uh, deceleration in progress, that even if you're in the Nigerian market, you're gonna want a pretty capable model. and given how cheap models have gotten, uh, are you going to tolerate a 2-year-old model, uh, if it's a tiny bit cheaper, or are you going to want something pretty close to cutting edge? I, I think the capabilities, uh, race is still on. And so unless Deepseek constantly offers lower prices and at the same time retains cutting edge capabilities, which is very hard to do, unless it does that, I'm not sure it's got a guaranteed pathway to dominating the Nigerian market, for example.

Speaker B: We were talking about, you know, being in this race and, and how these leads can evaporate so quickly, uh, you know, on, on the algorithmic side. How has that changed how you think about how valuable some of these model businesses are? How you think about, you know, AI business models, uh, writ large? Like, have you been sort of going down that idea maze a little bit? Speaker A: Well, I think if you look at the history of, of software in general, I think you, you have to conclude that although it's very difficult to measure software quality in any sort of durable manner, the, the key to success in selling software is not really software quality.

It is part of the— is one of the equations, but the companies that have durable businesses have durability not because they've got a quality advantage generally, but because they have a distribution advantage or they've got network effects that work to their advantage. I think if you ask yourself, does everyone use Google versus Bing because Google search is dramatically better? I anecdotally have a sense that Google's better than Bing. Could I quantify that for you? No. And do I think, um, Google is better enough to justify my using it 99% of the time over Bing?

No, I don't think so. And so I think when you start looking at the history of software, you find that actually distribution matters hugely. Network effects matter hugely. And so when I think about AI business models, the question I would ask is, is, is, is not who's got the best model this year or who will have it next year, but who's got consumer attention, who's got enterprise attention, and who's got the ability to keep that attention over the long run. And there, I think, asking yourself, what are the, what are the best metrics for assessing which model company, for example, will best win in the enterprise?

You know, performance on this or that benchmark is probably not going to be the primary variable. You don't want to be far, far behind in performance, but I think questions of quality will matter a lot. Your ability to not hallucinate, especially not hallucinate in the wrong ways, costly ways. Your ability to manage customer relationships will matter a lot. Your ability to bundle certain offerings together in a way that make it easier for enterprises to buy, that'll matter a lot. And all those are things that are not at all measured by benchmarks.

And so I, I find myself asking questions about that a lot more than I do about who's going to be ahead. Will it be Groq 3 or OpenAI in 6 months' time? Speaker B: I mean, from that perspective, it feels like, uh, very much OpenAI's to lose, uh, given, given the extent of the brand value they've already accumulated. One other sort of way I could see it playing out that I'd be curious for your take is it does feel like Anthropic has done a really good job sort of appealing more and more to an enterprise client base.

Speaker C: You know, there's some. Speaker B: Some interesting ways that I think they've tailored their product that, that feel like they can really have a foothold there. You know, among the, the large, you know, US players particularly, do you start to see where one moves and the other sort of fills in a gap or, you know, uh, takes over a bit of territory that's left undefended? Speaker A: Yeah, I think I would agree with you that ChatGPT certainly has the vast majority of consumer mindshare. and also agree that Anthropic, uh, I, I would say punches above its weight in, in enterprise.

I think that the big question to me is, as we start seeing more applications built on top of models, A, what models are those built on top of in the long run? B, how switchable are they? Um, and C, what's the, what's the business model of that, um, over time? Um, you know, I think most people probably myself included, had you asked me 15 years ago, uh, would be really surprised at how good of a business cloud computing is. Um, I probably would have said, well, it doesn't sound that hard.

It sounds super capital intensive. Um, margins surely can't be 50%, you know, permanently far in the future. And yet it has been. And so I think one of the key questions is, will you have models that are able to deliver that type of margin profile with others built on top of them? And right now that's a real open question. And I certainly see a lot of, a lot of data points of companies very deliberately trying to be able to switch between models to not have to pay that type of price.

Speaker B: Do you think meaningful value will be created at the application layer too? You know, you talked about the importance of this productization. I think within VC, I'm sure you've, you know, tracked this conversation, but we've really gone through various whiplashes around it where, you know, the value is in the infrastructure, the value is in the model layer, the value is You know, software's dead, apps are dead, wrappers are dead. Now wrappers are, are back. Um, yeah. How have you sort of, how do you see that at the moment?

Speaker A: You know, I think if, if you believe that, um, moats in software are less about capability per se and more about the ability to market capability, to productize, um, to have the right distribution channels, I think that suggests that actually applications have a real future because you can tailor to different end customers. And so I think I've always been on the side of— I suspect there will be durable business models for at least a couple of model providers, but not having to pay for the infrastructure, just sitting on top of it seems like a pretty attractive place to be.

Speaker B: We've, we've talked a little bit about, you know, some of obviously the work being done in China and the US. I'm I'm curious for your take on sort of the movement around sovereign models. Uh, you know, there's been a little bit of conversation around the idea that these are sort of a new type of strategic asset and that every major superpower or, uh, superpower-to-be will need to control some, some model of their own almost. And so you have, uh, Sarvam in India, you have Sakana in Japan, you have Mistral in, in Europe.

Is that a, a view of the world that, that makes sense to you? Do you think that's how this, you know, potentially plays out? Speaker A: I think the question to me is what problem you're trying to solve with sovereign AI, what that problem actually is. I, I think there's a, a fair number of, uh, cases of sovereign AI, uh, where the problem to be solved is, you know, politician A hears AI is important, wants AI in his country. doesn't know what that means. And so build sovereignty. I think there's a fair amount of, of that happening.

Speaker B: Is that a bad idea though? Well, I, I think that seems like it might be a really good strategy as a, as a president of a country. Speaker A: Well, I, I, good strategy for, uh, um, for photo ops or for, uh, long run, um, um, capabilities for your tech sector. I, I think if, can a medium-sized country really compete against US tech giants in the long run? I doubt it. I doubt it. Just looking at the scale of infrastructure that's being built right now, um, looking at the expertise that's required, it doesn't seem to me plausible that you're going to have a, a medium-sized European country or a medium-sized Southeast Asian country spending, uh, many billions of dollars a year replicating capabilities or probably actually replicating slightly worse capabilities, um, than you can, uh, buy for almost free from Silicon Valley.

Now, do you, you gain some sort of sovereignty from that? Well, you know, what does that mean exactly? Can you get the model to spit out what you'd like to spit out? Yes. How much are you willing to pay, uh, to have, uh, a model that spits out what you want it to spit out versus regulating access to ChatGPT or other models in your country, which, uh, countries already know how to do. So I, I, my sense is that if you're worried about what models say, it's much easier and much cheaper just to regulate access to them in your country.

And if you're asking yourself, is there an economic case— not a political case, but an economic case— for having your own model, I would say probably not if there's a fair number of roughly comparable in capability models that you can already access and don't have to pay the startup costs for. Um, so I, I am a bit of a skeptic of, of sovereign AI. Could I be dissuaded in certain quite specific use cases, I think I, I could be. But right now, it— I don't know that I've seen a great example of a sovereign AI project that looks smarter than imposing whatever national regulation you'd like to impose on US model companies when they offer products in your country.

Speaker B: More, more productive would be investing in, you know, perhaps institutes that, that teach people to bring on board these skills, you know, coax some of these large companies to open offices there. That's sort of the, the path that, that feels more, more useful. Speaker A: I think if you're, if you're looking at economic value, that is by far the path that looks, uh, the most useful. And then it's just a question of how much are you willing to pay for addressing your political concerns and what's the right way to address your political concerns.

And it just doesn't seem to me that in anything but the most extreme cases, and China is probably the extreme case, it would be worth the cost to try to replicate less effectively what Silicon Valley is already doing. Speaker C: This episode is brought to you by Explode, the best way to share data, dashboards, and reports with your customers. Are your customers asking for more data and reporting? Are you tired of manually pulling data or having your engineering team build analytics? You've probably gotten these requests, but they drag down your team with work that isn't the core focus of your business.

Enter Explo, an all-in-one platform for sharing data and analytics with your customers. Getting started is easy with Explo's low-code functionality. You can build dashboards, emails, or set up data exports for your customers. Once you're ready, simply add dashboards into your SaaS application or start sending data directly in the Explo platform. The best part? Your users Users can use Explo's AI features to create their own reports and dashboards, eliminating data requests for your support team. Simplify the way you share data and analytics. Try it free at co/mario. That's co/mario. Speaker B: We've talked a lot about the technological differences between the US and Chinese ecosystems, but a little bit less about the political undercurrents.

Perhaps you could tell us how you see, uh, the Trump administration's role in this technological battle and, and sort of the new territory, uh, the new struggle of the chip war. Speaker A: I think there are a fair number of, um, points of continuity between both the Biden and Trump administration and also the Biden administration and the first Trump administration. I, point one would be the focus on having more diversified semiconductor manufacturing supply chains. Uh, TSMC recently said that they produce 99% of the world's GPUs, all of which were produced in Taiwan, uh, which is an extraordinary, uh, figure, an extraordinary level of concentration.

I think actually you'd struggle in history to find another moment in which the, the key ingredient to technological progress was produced only in one country by one company in just a couple of factories. And so I think there's, there's broad agreement in the US political system about the desire for more diversification, including having some production, uh, in, in the, or at least somewhere outside of East Asia. So that's one. I think second, uh, the basic, um, belief that the S. will be better off if it leads in AI. Trump has called this AI dominance.

The Biden administration didn't have a, uh, the same phrase, but they had the same concept behind, uh, what they were trying, uh, to pursue in terms of cutting China off from accessing high-end chips and, and making sure it was only S. and allied firms that that had them. I think there's a key difference around what the Biden team called AI safety, and which I think the Trump team looks at as efforts to slow down AI or to make it, um, uh, politically biased. And so that will, I think, be a major shift, uh, between Biden and Trump.

Um, but I think it was notable the representation of, uh, big tech CEOs at Trump's inauguration, uh, the extent to which, um, Silicon Valley and Washington see themselves as not aligned on everything, but I think there's a fair amount of alignment at least on the AI race and how important important it is, uh, to win it. Uh, and that's something that would have been expressed in different language by Biden and Democrats, but I think the, the core dynamic of there being a race and it, uh, being an important race is, is something that, that will be broadly shared.

Speaker B: You talked about the US's need to find new places outside of Taiwan to create these GPUs, and, and there have been some large initiatives, I think in Arizona and elsewhere to try and onshore some of that. And you also mentioned, you know, maybe outside the States and outside of Taiwan. How significant have some of those moves been? And how far are we from there being, you know, meaningful fabs outside of Taiwan? Speaker A: So I think there's some good news and some bad news on this front. The good news is that TSMC, the Taiwan Semiconductor Manufacturing Company, currently has up and running in Arizona a plant that is one generation behind the cutting edge.

And will, I think, be upgraded in the future to be closer to the cutting edge and will be supplemented by 2 more plants, uh, before the end of the decade. So that is not going to make the U. S. anywhere close to independent in chipmaking, but it is a big change from the past when TSMC had basically no sizable presence outside of Taiwan or China. So that, that's a positive change. And I think it does provide a base that can grow. I think the, the negative change is that the leading US chipmaker Intel finds itself in an even more difficult position today than it was in a couple years ago.

And Intel is in the midst of a CEO search as we speak. There's widespread public discussion in the media of different ways the company might be split to make it more efficient. There's huge uncertainty that hangs over Intel, and I think one of the challenges that not just the company, but all of Silicon Valley and also the Trump administration face are going to be what happens with Intel. Um, because right now the market for manufacturing advanced chips is, is just 3 companies. And one of those companies, TSMC, is, uh, by far the dominant player.

Both Intel and Samsung, the third company, uh, have really, really struggled in their chipmaking business in the last couple of years. And so neither Washington nor I think Silicon Valley want to see a situation in which there's only one player and it's located all in Taiwan. And so that will be a key question to watch the next couple of months, how that plays out. Speaker B: In preparing for this podcast, reached out to a few of your friends and colleagues. And, uh, one of the questions, uh, that was suggested I bring to you is not an easy one.

Uh, but, um, if you were given the task to try and save Intel, uh, what would, what would your strategy be? What do you think makes sense from here? Speaker A: You know, I think the key challenge that Intel has faced for the last couple of years and has not resolved is who will be the customers for the new chipmaking facilities it plans to open, uh, over the next couple of years. So Intel's got in-house a chip design business that produces chips for PCs, for example, and many of those chips will be produced in-house.

But Intel has been trying to win external customers for its manufacturing operations and thus far has not succeeded at scale. But everyone agrees that because of the economies of scale in the industry, it's got to win significant volumes of external customers to be viable. And so the question is, who will these external customers be? And there's only a small number of companies in the chip industry that have the volume, uh, necessary to contract with Intel to manufacture chips for them. NVIDIA, Broadcom, Qualcomm, Apple, uh, they're all Silicon Valley firms, and there's only a half dozen of them, maybe, maybe 10 at the highest, at the upper levels.

And so all of those companies, I think, have almost certainly been talking to Intel, trying to understand their manufacturing process. But from what we know publicly, none of them have been willing to make a big bet on Intel. And so the key question that's got to be sorted out is who will the customers be that will provide the volume that lets Intel scale its manufacturing? Until you can answer that question, I think there will be a very big, uh, uh, uh, uh, there'll be a lot of uncertainty hanging over the company's future.

Speaker B: You lay out in an extremely vivid way in Chip War the convoluted, concentrated bottleneck supply chain around semiconductors, uh, you know, not least with companies like ASML and, and how, you know, just to create their machine, which is used in the process, they have to, you know, find the most magnificent mirrors and lasers and all these other sorts of things. As you think about the way the supply chain is, is changing, what are you looking out for? I know that you've written before about, you know, potentially some of the countries in Southeast Asia playing a larger role.

Uh, yeah. What, what's important to, to watch here? Speaker A: I would focus on two, uh, different dynamics. I think one is, is China. The second is Southeast Asia and India. On, on the Chinese side, China's historically been the world's largest importer of semiconductors. It hasn't produced large volumes of chips domestically. It imports them from Taiwan, Korea, and elsewhere. But that's beginning to change. When it comes to low-end and mid-range chips in particular, uh, China is rapidly racing towards self-sufficiency. And that matters because, uh, one of the really only products, the only manufactured goods that China imports from the rest of the world is semiconductors.

China's self-sufficient in basically everything else, uh, when it comes to manufacturers. And so if you were to ask yourself, what are the things China needs from the outside world? What are the things that in a crisis China would struggle to do without? Today, chips are at the top of that list. In 5 years' time, chips will not, I don't think, be at the top of that list because China will be much, much more self-sufficient when it comes to, uh, mainstream semiconductors. And I think that's going to have huge geopolitical implications, as well as obviously meaning a lot for the companies that are involved.

And so China's success and self-sufficiency in mainstream semiconductors is something that we haven't thought enough about with huge implications for Taiwan. So that's, that's one big change in the supply chain. I think a second is the way that Southeast Asian countries like Malaysia, but also India are trying to play a much bigger role. They've got huge ambitions. They think that The effort by Western firms to shift their supply chains away from China open up real opportunities for them. They've got rising incomes, growing populations, especially in India. And so if you look at how Apple, for example, has started producing 15% of iPhones in India, both because it wants access to the Indian labor force for manufacturing and it wants better access to the Indian market for selling, I think that's a great data point for how tech supply chains in general are becoming more dependent on India and Southeast Asia.

Speaker B: When you say that in 5 years' time, China will have, you know, will no longer have chips as the highest on its to-do list to sort of bring in-house. Do you, do you mean that even in the sense of those cutting-edge chips, or is it sort of everything up to the latest, you know, 2 developments, how, how far behind, uh, are they and, and how close can they get in a 5-year period? Speaker A: So I think we can say with confidence that, um, already China's got all of the know-how needed to produce everything up to 14 nanometer with pretty high degrees of efficiency.

And over the last couple of years, China's bought more, uh, quantities of chipmaking tools than ever before. Not just for itself, but actually any country in history, China's bought more. So they, they're building out the manufacturing base for everything that's 14 nanometers and above, which is, those are exactly the types of chips you need for, uh, household appliances and automobiles and vehicles, but they're not the types of chips you need for AI, for smartphone processors, the, the really high-end chips. That's where there's still uncertainty right now. China can produce small volumes.

Of 7-nanometer chips, which are themselves a couple years behind where TSMC is. And they're producing some for Huawei smartphones, some for Huawei GPUs, but we know that Huawei is capacity constrained when it comes to GPUs because Chinese firms are still importing large volumes of NVIDIA chips that are legally downgraded, uh, to be allowed to sell into China. In the future, China is going to try to use every possible way to break through the barriers that the S. has set up. And so my guess would be that in 5 years, China is still behind TSMC, but there is uncertainty there.

Um, and if China were to catch up with TSMC, it would have really profound implications. Speaker B: To catch up to S. capabilities and TSMC's capabilities, is your expectation that China has to essentially create an alternative to an ASML, ASML, like EUV machine? Is it that they, I mean, these are not super portable machines, but is it that they somehow managed to, you know, get one into the country in, in some way, shape, or form? How does that happen? Speaker A: So these machines are far too big to smuggle. Uh, there aren't any sitting around unused, uh, in the world.

And so the question is, can China either A, replicate everything ASML has done, or B, find an alternative way of producing very small transistors on its chips. In terms of replication, we know there's been a number of publicized incidents of employees leaving ASML, going to China, facing allegations of IP theft. I presume that the Chinese government has hacked in or tried to hack into every major semiconductor company. But, uh, we've now had 6 years of prohibition on the sale of ASML tools to China and basically no evidence, uh, whatsoever that China's made progress in replicating an EUV tool.

So this is extraordinarily hard to do. I think it's worth Following the second option, which is that China finds an alternative way around that doesn't require an EUV tool. Now, the hard part here is that no one knows what this would look like, um, because it hasn't been done. And of course, there have been a lot of smart people with a lot of financial incentive to find cheaper, different ways to make high-end chips. But it's also worth tracking this closely as well, um, because if you can't replicate an ASML tool, perhaps it's easier just to find an alternative pathway.

Speaker B: We've started to see startups enter the chip space. Obviously some of the larger companies are making chips of their own. Are there particular trends or new design paradigms that you're following that you think are particularly promising or you could see there being an avenue for one of these new companies to really get a meaningful hold? Speaker A: I think one of the big trends in the industry for the last couple of years has been what's generally referred to as advanced packaging, which means bringing different types of chips together in different ways, in formats that allow for faster data interconnect, for example, or help address heat dissipation issues.

And a lot of the capabilities that you need for better advanced packaging aren't necessarily about creating the smallest transistors. They're about using materials in, in different ways, for example. So they leverage different types of expertise. Now, I think in the Western industry, most people would see advanced packaging as a complement to having access to UV lithography. But I think it's worth being sensitive to the opportunity that you could have some pretty important breakthroughs that would provide you an entirely different pathway for improvements that take advantage of material science, take advantage of capabilities in chemistry and physics that are focused on advanced packaging rather than the optics-focused pathway that ASML's lithography requires.

Speaker B: The AI revolution has, among other things, also increased the demand we're likely to need energy-wise. How have you been starting to think about that part of this supply chain? You know, I've spent some time studying companies like Commonwealth Fusion Systems and, you know, think that's a really fascinating way of trying to address our need for limitless energy. What are you paying attention to? Speaker B: The AI revolution has, among other things, also increased the demand we're likely to need energy-wise. How have you been starting to think about that part of this supply chain?

You know, I've spent some time studying companies like Commonwealth Fusion Systems and, you know, think that's a really fascinating way of trying to address our need for limitless energy. What are you paying attention to? Speaker A: So I think about this in terms of short-term versus long-term. I think in the, the longer term, whether it's fusion or, or small modular reactors, there's a number of interesting capabilities that would really be transformative and provide the power that will be necessary. But I think for the next 5 or 7 years, we'll be operating in a world where you have to make do with some combination of solar, wind, gas.

and maybe if you're lucky, some, some hydro or other types of renewable powers, but only in certain places. And there's not that much unused hydro sitting, uh, around. And so I think for the, the short run question is, can we find enough power and can we build enough of the infrastructure, all of the transformers, have the gas turbines in place and do it fast enough to meet AI infrastructure demand? Because in the short run, I don't think any of the key trend lines are going to meaningfully. Change. And the key dynamic has been this, that for the amount of computing power per dollar of spend has doubled every 2 years, um, in the machine learning era.

And the efficiency measured by watts per flop has doubled every 3 years. And so you just project a doubling every 2 years versus a doubling every 3 years and use a ton more power. And that, that has been a relatively durable, uh, dynamic. I think we've got to try to change that in the long run, but it's not going to change in the short run. And so I, I've been struck by the number of folks I know at hyperscalers and AI labs who have become very deep experts in things like transformers, not because they wanted to be, because they had to be, because now the AI supply chain stretches all the way into power infrastructure.

And actually the limiting factor might not be electrons, but actually the transformers that you need to handle those electrons and deliver them to the data center at the right voltage required. And so that, that is a limiting factor that I think most people would not have expected a couple of years ago, but has turned out to be fairly binding. Speaker B: Are there ambitious projects in that space, you know, specifically that do address that sort of 5 to 7 year period that you think are really interesting and, and maybe play a meaningful role here?

Speaker A: You know, I think there's a limited set of options in terms of how you can produce, say, a gigawatt of power. You can do it with gas, you can do it with solar if you can find a solution for when the batteries are no longer full and the sun's off. So some sort of battery, gas, solar combination. You can do it with wind in certain places, but again, you have intermittency issues. The current energy mix only gives us so many options. And so I think that the question is, can you in a given location find the right amount of surplus power or bring enough power online in a time horizon that is viable.

It ends up being, I think, probably less about creativity and more about can you manage the permitting processes around, uh, grid interconnection or building a lot of power, which is certainly not the world in which, uh, AI companies wanted to be living. Um, but they're still constrained by those realities. Speaker B: Speaking of big infrastructure projects, uh, Stargate, uh, is— was certainly in the news not, not long before we've had this conversation. How significant is that as a project in your view? It certainly captured a lot of attention and, and, you know, is showing how willing these players are to invest heavily.

Uh, will it, will it make a meaningful difference in, in the way that this industry shapes out? Speaker A: I think we're still learning exactly what the contours of, of Stargate are. We know that OpenAI's got, um, a big data center complex, for example, in Texas, uh, under construction. That's been known for some time. I think the fact that SoftBank is going to put, um, a fair amount of money, it seems, into this effort is significant. We haven't in the past had a lot of venture funding going into infrastructure. That is a big new change.

And I think, uh, it's a positive change because you need a lot of money going into infrastructure to build out, uh, everything that's required. I think a lot of the headlines around Stargate spending $500 billion were shocked by the $500 billion figure. And I've got no direct intelligence as to whether that's the right final number, but I would just— yeah, but I would just observe Microsoft's going to spend $80 billion this year on AI infrastructure. And so the idea of a leading AI company spending $500 billion over 4 years is not that much higher than Microsoft's observed spend, uh, this year.

So, so even if you're a Stargate doubter and you think it's going to underperform, I would posit that is just one data point among many, that there is just a vast quantity of infrastructure spend already happening. Uh, and we, we should see Stargate, I think, in, in that light, even if it's not as real, um, as Sam Altman suggests. I think it still certainly speaks to Speaks to the fact that big AI companies realize they need a ton of infrastructure. Speaker B: Yeah, and maybe we don't see many on the order of $500 billion, but the number that we see in the $10 billion, $50 billion, you know, sort of segment probably increases over the next few years, I expect.

Speaker A: Well, in this year, Amazon has put a number of $100 billion of AI spend on the table. Facebook $60 or $65, I forget, billion. These are huge sums of money. Uh, and again, I think when you start listening to those numbers, $500 billion over 4 years doesn't sound crazy. Speaker B: Yeah, absolutely. Before you studied semiconductors, you spent a lot of your career studying Russia, uh, and, you know, the, the, the major characters there, including Putin. And in our previous conversation, uh, our Modern Meditations interview, you talked about how Putin is really one of those figures who has actually profoundly influenced how you see the world, uh, and in particular how studying him made you appreciate the different motivations that even very modern people have, uh, and that in, in his case, you know, he's, he's really motivated by very different things than what we might expect of a traditional leader.

I'm curious, as you look, uh, at the panoply of different AI magnates, uh, of, of different stripes, you know, you have the, the Sam Altmans, you have Elon Musk, Jensen Huang, uh, Mark Zuckerberg. How have you started to think about parsing their motivations and, uh, and how modern or or perhaps sort of ancient their drives might be. Speaker A: Well, I think certainly you're right that I've spent a lot of time thinking about Putin in the context of Russian history. The Russian Foreign Minister Sergei Lavrov somewhat famously said that Putin only has 3 advisors.

They're Peter the Great, Catherine the Great, and Ivan the Terrible, which I think is true to the extent that it illustrates that he's not maximizing GDP, he's not maximizing household income. He's not maximizing any of the things that you would be taught in Political Science 101 that political leaders often maximize. He's maximizing empire or glory or something different than that. I, I wouldn't want to be on the record as comparing, um, any of our Silicon Valley CEOs to Vladimir Putin. However, um, I think, I think the basic, uh, um, I think the insight that empire building, uh, still is a, still is a character trait, uh, that characterizes some of, um, some of our, our most influential business leaders.

I think that's, that's true. Um, another way of looking at it would be, you know, if you tried to understand any of the figures you mentioned, whether it's Mark Zuckerberg or Elon Musk, by simply maximizing the number of dollars that they have in their bank account, I think you'd probably fundamentally misunderstand what drives them at this point, both because the, at some point, the marginal utility of new dollars declines when you've got a lot of them. But also because I think they see their roles as being beyond that. And there's room, I think, for critique of their visions as well.

But yes, it does seem that if you believe you're at this moment of rapid technological change, you believe you're at this moment of fundamental disruption of, uh, not just the tech sector, but the entire corporate hierarchy that is going to be set off by this technological change. You're probably thinking about it not in terms of, uh, dollars and cents or, or the margin profile of your business per se, or, uh, anything, uh, that you might be taught in an accounting class. You're probably thinking about in terms of conquering, uh, and defeating, uh, your rivals, uh, as, as the first order approximation of, of, of what's driving you.

And then the, The financial metrics are sort of downstream of that more primal desire. Speaker B: Yes, probably way downstream for most of these folks, as you mentioned. Um, I'm not, I'm not sure if this is, we'll see where, if this goes anywhere, but I wonder if there's something unique about AI as a technology that appeals to a different set of motivations beyond empire building. And I suppose what I mean by that is. AI is so uniquely or singularly, uh, powerful, and it seems to at least speak to our desire to, you know, create God in one way or another.

Maybe that's too histrionic, but it's not so histrionic. Are we thinking of leaders that are considering apotheosis in, in some capacity in, in the motivations here, or am I, uh Am I simply searching for something grander than really exists? Speaker A: You know, that's an interesting question. I think we've, we've heard certainly Elon Musk talk about the, the need to harness AI so that humans are able to compete against AI in the future, that we have some sort of competitive edge. I think that's interesting to think through. I guess I, I'm, I'm struck by on the one hand, I think you're on to something.

On the other hand, the, the day to day of AI companies is of course, producing marginally better products that can, uh, win market share from, uh, their rivals and things like code generation, which has a certain parallelism to, uh, what you might do if you're trying to create God, but also seems pretty mundane when you're on the day-to-day of the running. Speaker B: Yeah, debugging your, your code stack somewhere, your, your code base. Uh, yes, that's probably very true. And in the day-to-day, maybe you get into that. conquering energy, and then on your retreats you get to pontificate a little bit more.

I'd love to finish up with a few modern meditation style questions, uh, and jump into some, some fresh ones that hopefully, uh, I don't think I asked you last time around. So here we go. Um, if you had unlimited resources and no operational constraints, what experiment would you want to run? Speaker A: So I think one of the social processes that is the most important, but which we understand the least, is drivers of economic growth. And it's impossible really to run experiments on economic growth. You can't impose different treatments on different countries.

But if you could, you would learn fascinating things from that. And the results could be really transformative to human welfare writ large. So that seems to me the, the question set. That we know the least about that would benefit from hypothetical experiments like this. Speaker B: What would a, a plausible version— well, not plausible because we're, we're out of the realm of plausible here, but like, what would a version of that look like to understand how you would set up the, the right, the right constraints, the right variables? Speaker A: Well, we've had a couple of historical experiments that were produced for us by history.

The division of North and South Korea, for example. Told us something about what works and, and what doesn't. Um, East and West Berlin, or East and West Germany, uh, being a second. And so you could imagine, um, in a hypothetical world, drawing random lines across countries and applying different policy sets, uh, to each of them and seeing what the results would be. Today you can only do that at very tiny scale, uh, or you can only do it by finding historical aberrations like countries that were divided. Speaker A: Well, we've had a couple of historical experiments that were produced for us by history.

The division of North and South Korea, for example. Told us something about what works and, and what doesn't. Um, East and West Berlin, or East and West Germany, uh, being a second. And so you could imagine, um, in a hypothetical world, drawing random lines across countries and applying different policy sets, uh, to each of them and seeing what the results would be. Today you can only do that at very tiny scale, uh, or you can only do it by finding historical aberrations like countries that were divided. Speaker B: I wonder how far away we are from being able to realistically simulate that with AI worlds.

Um, interesting thing. Speaker A: That's an interesting question. Speaker B: Uh, what's a practice we should borrow from another culture or era? As a historian, I suspect you maybe have, you know, a thousand of these and plenty that we, we shouldn't adopt. Speaker A: Well, I think the, the practice that I, uh, think the most about and ask myself how would I implement is back to the era of civilian scientists during the Renaissance, when you could just be a generally educated person and build yourself a telescope and look up into the stars or think about what causes things to fall to the ground.

I read about those scientists and, and really look at them with wonder because I feel like nothing that I do or the scientists that I know do is anywhere in the realm of exploration of the type that was done at that time. And I think it's fascinating to ask yourself, what are the types of questions that would have been asked by someone like Newton or Galileo that you ought to be asking yourself? And I guess I generally feel that I'm only asking questions that are an order of magnitude as interesting as some of those figures in the Renaissance.

Speaker B: What a great prompt for oneself on a maybe weekly basis. What would, what would Newton do? What would Newton be asking? I was only recently learning that Newton really, uh, sort of burned out his mind towards the end of his life and went down a long rabbit hole of biblical numerology. Speaker A: There's an extraordinary biography of Newton, um, by an author named James Gleick, a short book, uh, but really puts you in the mind of, of Newton as he's trying to make sense of the world around him.

I strongly recommend it. Speaker B: Oh, fascinating. Uh, what's an underappreciated corner of the internet that you think more people should know about? Speaker A: Where I try to spend more time than I actually do is, is reading foreign language newspapers, which is an analog thing to do, I know, on the internet. But I also find that if you want differentiated opinion that is curated for you, but pushes you beyond what you're, uh, reading in your English language echo chamber, that's, that's one of the best places to find it. Speaker B: What are your chosen non-English newspapers?

Is it, you know, are you perusing Le Monde or, you know, some of the— Speaker A: I, I am biased by the languages that I can, uh, make sense of, uh, so, so generally European newspapers plus the, the Russian press, which is of course, uh, full of falsehoods but also gets some things interestingly right, uh, and I learn a lot from reading the Russian media. Speaker B: I should have realized that you'd be trying to read it actually in these languages. I was just assuming you were going somewhere and, you know, doing a nice Google Translate on each of them.

Speaker B: I should have realized that you'd be trying to read it actually in these languages. I was just assuming you were going somewhere and, you know, doing a nice Google Translate on each of them. Speaker A: Sometimes too. Speaker B: Final question. What was the last great piece of media that you consumed? And it could have been, you know, a film, a book, an article. I'm asking this extremely selfishly so that I might check it out. Speaker C: After this? Speaker A: There's a wonderful book just published last year called A Brief History of Intelligence, actually written by someone who is not a, a neuroscientist, uh, or an academic by training, but wanted to understand how intelligence had developed from the first brains, uh, over many hundreds of millions of years of evolution.

Uh, I thought it was highly illuminating, learned a ton about how the brain works, and learned a lot about how the brain has changed, which I'd never really been exposed to before, and I think it was particularly interesting to think about how are our artificial brains, artificial intelligence systems changing over time, and are we evolving them in a similar way that human brains evolved? And the answer is yes in some ways, but I think no in others. So it was immensely thought-provoking to read. Speaker B: Wow. Incredible. Chris, thank you so much for joining us.

Speaker A: Well, thank you for having me. Speaker B: That's it. Speaker C: Thank you for listening to this episode of The Generalist Podcast. Please subscribe on Apple Podcasts, Spotify, or your preferred podcast app. Ratings and reviews help others discover these discussions, so if you enjoyed the conversation, I'd be grateful if you could take a moment to leave one. For all past episodes and more, visit us at com. See you next time as we continue to explore The future. Speaker A: Well, thank you for having me. Speaker B: That's it.

Speaker C: Thank you for listening to this episode of The Generalist Podcast. Please subscribe on Apple Podcasts, Spotify, or your preferred podcast app. Ratings and reviews help others discover these discussions, so if you enjoyed the conversation, I'd be grateful if you could take a moment to leave one. For all past episodes and more, visit us at com. See you next time as we continue to explore The future.