Tyler Cowen (Part 1)

26:54It worked out for him until he went crazy. Over the last year, biggest political changes and upheavals? Well, the Middle East, everything that happened with Iran and continuing conflict in Gaza, I never know how significant that is. Like, that was my baseline forecast anyway. I guess what I've seen is that what I expected has happened. Sort of an anti-surprise. But sometimes the anti-surprise is surprise. Like, whoa, I was right about that. What a surprise, you know? Also, if what you believe is extreme, seeing it happen is surprising. That's right. But I was not surprised by any of that. Not surprised by Russia, Ukraine.

27:40So those are two big areas where I feel I predicted mentally very well. And we're just seeing happen what I expected. And I suppose China, Taiwan will see how that turns out. But I'm pessimistic. But it hasn't happened yet, right? Tell me about the betting market, what you can learn from looking at the betting market in general versus what the experts say. The betting market slightly outperforms the experts. It depends who you take to be the experts. Ex post, you can always find an expert who got it right. But ex ante, I think when you have cases where there's genuinely no information to go on, the experts might be a little bit better than the betting market.

28:25They just have some human intuition. But those are not many cases. So when there is this rumor recently, Trump was very sick, he might be dead. The betting market gave that, I think, a bit more credence than the experts. Experts maybe were right there. He still might be quite sick. We don't know. But there's really nothing to go on. And then I don't think the betting market adds much value. It might be a tie with the experts. But when you have data, things like who's going to win the Super Bowl, what's the best NBA team, I mean, you just look at the betting market and you're not going to do any better than that.

29:02It's just the best source. Presidential elections, maybe the experts in betting markets are tied, but they're tied in part because the experts are looking at the betting market. So they're also interdependent. but it's just a good way to find out probabilities and you shouldn't take it too seriously but it should disabuse you of notions to the contrary and then it's like, well, if you're so sure, make a bet so there's people who tell me there won't be another election in the US you know, fascism will descend on us I go, I look at the betting market last time I looked, Democratic Party was 53 % likely to win the next election okay, you might think that's wrong but make your bet, you know.

29:4453 % sounds like a normal election to me. You don't have to think the candidates will be normal or everything's fine with the country. Clearly it's not. But I strongly believe we're going to have another election and whoever wins wins and then we'll have an election after that. Are there any things in particular that you would look to the betting market and think this is the best place to get information? Well, anything with regular data. It's probably the best place. And sports, maybe they're not a socially important example, but they're a very clear example. And there's a Las Vegas set of betting odds, prediction markets.

30:19They're basically telling you the same. Because of transactions costs, they can deviate a little. What's another example beyond sports? Well, anything with elections. Or, you know, will a certain country leave the EU? Right now, it doesn't seem to be in the cards. But if I were wondering about that, I would look at a betting market, prediction market. and I don't think there'd be a better estimate. Who do you think are the people who are betting? Well, it's more open now, so it's easier for Americans to bet. You don't have to go through a VPN or a foreign credit card. So a lot more Americans. There's some insider trading in those markets that people who know go and make a bet, so you see the odds move before the news is announced.

31:07I just think you have a lot of somewhat wealthy people. who find it fun and attest their knowledge. Then you have a lot of people who are not that wealthy, and it's a form of gambling. But maybe the odds there are better than some of the other ways they might gamble. And they do it like they bet on sports. Tell me about the book Goat and the idea of it being a generative book. Goat is a book I wrote published two years ago, and it's the first book I know of published in GPT. then it was GPT-4. So what does that mean? Goat is just a regular book. You can read it in PDF or print it out like any other book.

31:47Put it on your Kindle. But instead of reading Goat, you can just ask the AI about Goat. And Goat has been fed to the AI, now also Claude, the other models, Gemini, Meta. And you can just say, I don't want to read Chapter 2. Summarize Chapter 2 for me. Or, oh, in Chapter 2, Tyler didn't explain where Milton Friedman went to school. tell me all about Milton Friedman in college, and it will do that. So it fills in parts of the book that are missing. You can ask, where would Paul Krugman disagree with this sentence? And it will give you a sense of that. So it's an interactive book. Any reader can build his or her own book by interacting with the AI.

32:25And I think books in the future will be like GOAT. There'll be the plain version you might just read, and then the AI version, and a lot of people will prefer the AI version. How will you feel if the AI's interpretation of your work is maybe even the opposite of what you meant? I might feel the AI was right and I was wrong. That would probably be the metarational thing to feel. I think I would be more flattered than irritated. The AI has a lot of opinions about me because I've written online every day for 23 years, and I've done hundreds of podcasts, as you have, and they're on different topics, so its model of me is pretty broad.

33:07And anything the AI says about me, I would just take very seriously. And this is already the case, right? It's not a hypothetical. When did you start the Marginal Revolution blog? It's about 23 years ago, I think, almost to this day. It was like early September, I think. How did you get the name? Hmm. Marginal revolution refers to a movement that built modern economics to understand that people value goods at the margin, made possible the application of calculus to economic reasoning. But it was intended as a double meaning, that the blog itself would be a marginal revolution in thought. That's how most people took it, and that's fine.

33:48In a way, that's the better meaning. But it signals to an economist that we're economists, and that's good too. So I thought it was the perfect name. When you started it, did you have any prediction about how long you would continue it? I was totally wrong. I thought it would maximum get like 5 ,000 readers and maybe do it for a few years, and who knows. My current view is I'll do it until I can't anymore, that I'll never stop. You know, it would commonly get 50 ,000 to 100 ,000 readers. Plus, people read excerpts of it on different things where it doesn't show up on the counter, Twitter or other places.

34:23So it's been successful. I see no reason to stop. And you write it every day? Every single day. Now, I publish every day. I don't literally write every day. There's auto-publish. But I write like 95 % of days. Do you think of it as a sort of a diary? It was intended as a diary. It is a diary. I thought, I'll do this thing and keep track of what I've read, what I've thought. and the first year or two, that works fine. But when you have 23 years, I mean, there is a search function which has gotten better over time. But now a great thing about AI is if I want to find something on the blog and I don't remember the keywords, I just type into like perplexity.

35:05Where did Tyler write about this? And perplexity, GPT, Claude can find it. So it's a diary again in a much stronger way because of AI.

35:28LMNT Element Electrolytes Have you ever felt dehydrated after an intense workout or a long day in the sun? Do you want to maximize your endurance and feel your best? Add Element Electrolytes to your daily routine. Perform better and sleep deeper. Improve your cognitive function. Experience an increase in steady energy with fewer headaches and fewer muscle cramps. Element electrolytes. Drink it in the sauna.

36:05Refreshing flavors include grapefruit, citrus, watermelon, and chocolate salt. formulated with the perfect balance of sodium, potassium, and magnesium to keep you hydrated and energized throughout the day. These minerals help conduct the electricity that powers your nervous system so you can perform at your very best. Element electrolytes are sugar-free, keto-friendly, and great tasting. Minerals are the stuff of life. So visit drinklmnt.com slash tetra. And stay salty with Element Electrolyte. LMNT.

36:55Tell me about how the blogosphere has changed since you started till now. Most of the blogosphere has disappeared. So people do Substack, which is a great company, a great service. They're good defenders of free speech. We don't do Substack for two reasons. One is just our archives are so large. It would be very hard to transfer them, and we want to keep our archives. But also, Substack is, I think, best for longish essays, 1 ,000 to 2 ,000 words that you send out periodically. And if we do four to five posts a day, some of them quite short, no one wants four to five emails a day. but bundling them together isn't quite right either.

37:37So WordPress works better for that than Substack. So we still do WordPress, even though the technology now would be considered pretty backward. Do you remember what the inspiration to start it was? I just thought blogging would be a big thing and should do a blog. There wasn't like a really good economics blog at that time, and we did it with my co-blogger, Alex Tabarrok. Were there other blogs that you were reading at that time that made you feel like I wanted to join this party? Yeah, I wrote a little for a legal blog called Volokh Conspiracy. I wrote some economics for them. Brad DeLong had an economics blog.

38:12A woman whose pen name then was Jane Galt had what was often an economics blog. And those were valuable. And I drew some inspiration from them, but just thought I could do more. Do you read the comments? Yes. Now, not every comment over time, But if a comment is put up in like the first two-thirds of the day, probably I'll read it. And do you ever find anything interesting reading the comments? Every day, always. Really? Absolutely. Now, there's crazy people in there, and there's very negative, destructive people, and people who insult others. But like all forms of social feedback are valuable, and I enjoy it.

38:53And some of them are funny. Some of them you learn things, like the commenter called Schuer, whoever he is, he's a super smart guy. He knows a lot about the healthcare system. So yeah, read the comments. How important is free speech? I'm a First Amendment absolutist. I think even libel law is too strict. It's certainly too strict in the United Kingdom. And what the UK is doing, like arresting 30 people a day, I just think is evil and completely self-destructive. apart from what it does to their own people. They're a laughingstock in the world now. And it's just someone needs to shake them and say, hey, wake up, this is a mistake.

39:35Even if you think these people are wrong in what they write, like a big chunk of the U.S. is laughing at you and it's harming your own reputation. What percentage of your blogs are topical versus just a timeless thought? I'm never sure what the distinction is. So it's whatever I'm thinking about, and that's driven a lot by the news cycle. But if I just read a novel or hear some music, there's this woman I've been listening to. It's Emily Linga, and she's 17 years old, and she lives in Dubai, and she sings and plays the piano. And it's quote-unquote amuletre music, and it's all covers, but she captures something about some of the original tunes that works for me.

40:23Right. And I'm going to post about her. Is that, like, timeless? Is that topical? It's sort of both and neither, right? Mm-hmm. Do you ever read about philosophical themes? Absolutely. All the time. Yeah. And a lot of my published academic work is in philosophy journals, not just economics. What do you think in today's culture is overrated and what's underrated? That we're shifting to an oral culture. People are starting to notice this, but it's still underrated. How far it's gone. It's underrated how much, say, the sources of authority that were around in 2000. I mean, you mentioned WEF before.

41:02There's just so many things like that. They still exist. Well-known people go to them. They do a thing that is a thing. They appear functional. They have funding. But I think they're pretty... Mean less and less over time, you think? Yeah, irrelevant. This is a common observation, but I think it's still underrated how far that's gone. And in terms of the audio portion, you think in terms of speaking to and listening to your chatbot and listening to audiobooks versus reading books, is that what you mean? Underrated how far that's gone. And YouTube, and of course TikTok. And you think it'll continue in that direction, away from reading?

41:42Yes. This is not entirely a good thing, because there's something about reading that's quite analytical, and you have a more reasoned back and forth. and I think you're more likely to converge on truth. But oral culture excites people, motivates them, involves them. And all the millions of people, say, who use ChatGPT as therapist, I think it's good for most of them, or at least neutral. You always have your suicide cases or nutters. But I think it's a big net plus. And it's oral culture. It's part of it. What is mood affiliation? The idea that you judge ideas by whether or not you agree with their mood.

42:20So there are people who are like optimistic or pessimistic, and they see everything through that lens. I think that's a mistake. Maybe oral culture encourages that a bit, because an oral presentation, it has a mood, right? You come out with some level of energy. Is this a sad song? Is this a happy song? A tragic song? Whatever. Print culture, it's colder, as McLuhan said, and that can be useful. You see this so much with Trump. People have moods toward Trump, and whatever you think the right mood is, I would say try to get yourself out of that mood and just think about the thing and analyze it.

42:56And just try that on for a while. And try not to talk about Trump at all. Just talk about the thing. And you can later switch into talking about Trump, but make a big effort to segregate the two. Do you think in general people are more or less informed today than 20 years ago? They have more information. They have more misinformation also. So it depends how you weight that. They believe more false things, but they know more, too. What about 50 years ago? I think same basic trend, but all the more so. The information diet was pretty impoverished 50 years ago, but that did encourage some shifts into quality.

43:37So maybe very lately classic novels are making a comeback, but I feel they were more important 50 years ago than they were 20 years ago. Yeah, we'll see. but print culture 50 years ago was really very important and that had some big pluses how about a hundred years ago that's 1925 there were great classic works being written then but the overwhelming majority of books seem unreadable and stupid to the modern reader whereas they do not say from 1978 a history book from 1978 can just be a very good or great book but there were not enough quality checks or standards back then. So sometimes people overrate those earlier eras.

44:21But you do have these lone geniuses, say like Nietzsche in philosophy, where today you don't have anything like that. People who will still be worth reading. There's very few of those today. People are more specialized. They carve out these niches. So in the 1920s, like you think what was written in fiction. You have Proust, you have Thomas Mann, you have Rilke, like unbelievable works that were not quite matching today. What do we learn from fiction that we don't learn from any other place? I think what we learn from fiction we cannot express through means other than fiction. Otherwise, it would not be a thing that we learn from fiction.

45:03But they give us richer models of human beings. They recreate parts of past history for us that historians cannot. They challenge our values. They're pretty wonderful and amazing. I'm doing a reread now of Thomas Mann's Magic Mountain, and it's just been fantastic. That is a work from the 1920s. Maybe it's 24. And the relevance of it to the current world, you mentioned World Economic Forum before, but it's set in Davos, which is an amazing irony, because all these people show up, and then they do nothing. And some of them are mentally ill and they're all spiritually weak or out of date. And you can read it.

45:42It's a parable about 2025. And the novel is really very fresh, whether or not you agree with all the perspectives. It's a phenomenal book to reread and that the location is Davos. It's so perfect. What's the benefit of challenging our values? You hope for improvement, but that aside, I just think it's part of what a human should do for intrinsic reasons is to challenge things, most of all your own values. Maybe sometimes it depresses you or you go in circles. I just still think it's a better way to live. Tell me about moving to the free press as a place to publish. Well, I wrote 10 years for New York Times.

46:19And in general, I was happy there. There's a lot you can object to with the Times, but it didn't infringe on me very much. I wrote for the Business Page, and they wanted me as a contrarian. I wrote eight years for Bloomberg, and I wanted a change of pace and I thought what Barry Weiss is doing is exciting and as I said in a piece I published for the Free Press I like the idea that they're a startup and they can fail and when you can fail things really matter and you see this in the history of music especially early years of what become great groups or performers they know they can fail it's tough doesn't matter if you're great and it so motivates them and sharpens them and to have a version of this in media I thought would be fun and good for me.

47:03Do you think of yourself as contrarian? I don't know what that word means. I guess I don't. It's non-contrarian to think of yourself as contrarian. So somehow we should try to escape those categories, I would say. In your exotic travels, how do you find the best places to eat? Well, this is a view I have where I disagree with what a lot of people say. If you are wise and careful, there will be an inverse relationship between the price you pay and the quality of the food. So a lot of the best food in the world is extremely cheap, and it can be in places like India or China or Mexico. You have your best, most memorable meals outside the major cities and fancy restaurants.

47:46How do you go about finding those experiences? If I'm in Mexico, or I speak Spanish to be clear, what I do is I get a cab driver. you want a man who is between like 40 and 55 years old, old enough that he's been around, but not so old that he's jaded or just not tasting the food anymore, and pay him some money. It doesn't cost much to bring you to his favorite places. And you just have to convince him you mean it. So it's local wisdom, really, yeah. Men who have jobs in the transportation sector know all the places, and they know other men. You know, in the United States, you could ask a woman where to eat and you get a great answer, as good as what the men would tell you, probably better, in fact.

48:29But a lot of poorer societies, there's a lot more sexual segregation. Women are kept at home more. They may have jobs, but they're not working in the sense of traveling around all day the way a fireman or policeman or taxi driver would be. Those are male jobs. So you should ask men. And do you find the best places to stay that way as well? I just use GPT. Pre-GPT. Well, if you go before Google, I think you can just walk around and see. And I would do that, show up with nothing booked and do fine. Or even again, ask a cab driver. And the same for the most interesting things to see? You might just know already.

49:10Maybe the reason you went. That's the reason you went, yeah. So art museums and food and music would be my main interests. And I feel I go there already knowing. Of all the exotic places you've been, which are the least like our culture? Give me the best version, least like it best, and least like it worst. I went to Yemen, I think it was 1995 or 96, the strangest place I've ever been. So I stayed in the capital city, which is the richest place there, and I was in the best hotel in town, and no person in the hotel spoke English. All of the women, this is again in the capital city, not the countryside, they all wore full veil.

49:52Every man carried a dagger. And I'd say 80 % of the men, they chewed this leaf called cot, Q-A-T. It's a kind of narcotic. You're not totally stoned. It's like chewing coca leaves or something. I don't know. I never did it. And that's the whole society. It's the veils, the daggers, the no English, zero chain stores, zero brands. There were markets. Food could be great. medieval architecture, at that time all fully intact. I don't know if it still is. And then you could go. It was safe. It was the most incredible trip I ever took. So that was a positive one. Well, not positive for them. Their lives were terrible and poor.

50:30For you. I'm asking for you. Yeah. I would not wish that life on anyone. India, I find always fascinating, much less like America than China. I think of America and China as having a lot in common, in fact. America and India, not so much. And just the number of religions, languages, views, casts, everything makes your head spin. And there's so much drama there. Always interesting experiences. But it's tough with things like noise pollution. The hardest thing about going to India, I sometimes say, is crossing the street. There's so much traffic. You can't count on the traffic light. There are rules, but you don't know them.

51:12So if you tag along with other natives crossing the street and walk with them, that's the way to do it. But sometimes I'll pay for a taxi ride, and I just say to the taxi driver, take me across the street. And he thinks I'm crazy, but I have no other way of getting across. And it's worth the pittance that you pay. That's amazing. Yeah.

51:35In a world of artificial highs and harsh stimulants, there is something different. Something clean. something precise. Athletic nicotine. Not the primitive products found behind convenience store counters. Not the aggressive buzz that leaves you jittery. But a careful calibration of clean energy and focused clarity. Athletic nicotine. The lowest dose tobacco-free nicotine available. Made entirely in the USA. No artificial sweeteners. Just pure, purposeful elevation. Athletic nicotine is a performance nootropic. Athletic nicotine is a tool for shifting mindsets. Athletic nicotine is a partner in pursuit of excellence.

52:20Slow release. Low dose. Gradual lift. Sustained energy. Soft landing. Inspired results. Athletic nicotine. More focus. Less static. Athletic nicotine. More clarity. Less noise. Athletic nicotine. More accuracy, less anxiety. Athletic nicotine. From top athletes pushing their limits to artists pursuing their vision. Athletic nicotine offers the lift you've been looking for. Learn more at athleticnicotine.com slash tetra and experience next level performance with athletic nicotine. Warning, this product contains nicotine. Nicotine is an addictive chemical.

53:10any thoughts about Luigi Mangione or the assassination of other corporate leaders that are happening these days? Well, on Twitter today, I saw an ad. I don't think it's the real Luigi, but they put an image in, probably AI-generated, that looks just like him, and I think they're selling shirts. So it's a weird world. In many ways, the current state of the world is very disturbing, but rather than just freak out, I think we need to try to confront it in some constructive way and move on to where we're headed. What do you think the cause of it is? Well, I suspect he's mentally ill. I don't know much about him as an individual.

53:50I didn't read much of the articles, but I think he thinks it's justified. And again, I don't know much about that company. I would just say any large enterprise typically ends up doing some bad things, and you can blame the leader for those. I think there's some further level of maturity where you just say, this is not the right way to improve the world. But one needs to take seriously that it happened and realize whatever checks and balances were in place that would have stopped it from happening 20 years ago, maybe they're not so much there anymore. So there's this other story. As we're speaking, I think seven different AFD members running in local elections have died quite recently, all in one province of Germany, Nordrhein-Westfalen.

54:35How can it be? It could be coincidence. There's weird coincidences. Seven? It could be a serial killer who somehow does it in a way that law enforcement cannot track. I haven't read the stories. I don't know. But the very fact that you think it's possible it's a killer is what matters, whether or not it's a killer. We all think it might be a killer. And that's what matters for understanding today's world and how to get people into some new mind space where that is more unthinkable again, I think is a priority. But there's no mega project way of doing it. It's all small bricks and being positive about small improvements and building human bonds in a way it never feels like you're solving that problem.

55:19But maybe after a few decades, a whole bunch of your problems have become better. In Luigi's example, you talk about not knowing much, but believing maybe there's a mental issue. and you talked about there being an ad for shirts with Luigi in it. That's right. Let's talk about that part of it, the embrace of the character. My guess, I don't know much about the market for shirts. I'd never heard of the company. I can't remember their name. So odds are they're not a very well-known company. This is a kind of Hail Mary pass, like, hey, we'll try this for publicity. Otherwise, we have nowhere to go but down and maybe people will start talking about us.

56:01And they're an upstart, and there's some chance it works, right? Yeah. There's always so, so much. But I think the story of the next five years, other than politics, is that AI will start intersecting with more and more other parts of our life, and we need people with positive visions of how to make that work for us rather than against us. It's this idea of these small bricks, bit by bit, is how good things happen, and people will come out with grand plans and predictions. It will be tremendous. It will be awful. Probably all of that is misleading. It's up to us to decide how good we're going to try to make it and each do some part of that.

56:43And now that you're one of the 100 most influential people in AI, according to Time magazine, I'm one of the 100 most influential in philanthropy, by the way, so we can share that joke in a way, but Time was not crazy to pick you. You are highly influential on AI to audiences who don't have that many other sources. So I would take very seriously the fact that you won this designation. It was not a mistaken pick. And for you to do your part, so does people think about it better and more deeply? Because you will reach these people. What do you think about developing the imagination required to create better prompts?

57:23How do we do that? To allow AI to do its best work, it's about becoming good at prompting. How do we develop that skill? Well, I read yesterday that GPT has a new service, which I haven't looked at yet. You toggle to something and it improves your prompts for you. I haven't tried it. I don't know how good it is. I suspect it's a big help. Other AI services, you can just ask. So if you have a prompt for GPT, ask Claude. If you have a prompt for Claude, ask GPT. That right there is a big help. But just trial and error and experimentation, which is mostly fun. Again, keep in mind, you need to think literally.

58:02Some people say the prompt goes better when I speak the prompt, because when I talk, I'm a bit wordy, and I give more context than when I type. Not true for everyone. Some people are just laconic. Maybe I'm a bit that way. For a lot of people, they find that's useful, to speak the prompt. So maybe succinct isn't the best prompt. It's not, no. You want to be a bit run off on the mouth because that gives it the context it needs where you otherwise don't think it needs it. So you have to overcome your inner laconic self. I would not guess that. That's right. And it's really not like humans. I sometimes say it's like training a horse or a dog.

58:42Ultimately, it wants to obey, but simple requests are not always honored immediately either. there's a trick to learning how to do it. And once you know some tricks, you do much, much better. Just go up to some random dog and say, sit, beg. It's not going to do it. So people try this with the AI and they're like, oh, this AI is crap. But it's like just yelling at the dog to sit or beg, you know. It doesn't quite work that way. How do you feel about AI acting like a human in response instead of acting like what it is? I don't know what it is other than what it does. I prefer my AI fairly Spock-like.

59:21I don't want it to charm me. I don't want to use it as a therapist. I don't want those parts of myself to be interacting with the AI. And I'm very happy when it's just the facts, ma 'am. That's what they used to say on Dragnet. Yeah. But I know I'm not the typical user. People want all sorts of weird things from it, and they get it. I don't think we can fight that. I think we need to work toward a better version of it. You know, it might be interesting thinking about this. People like watching Twitch to watch other people play video games. It might be interesting to see and hear other people talk to their AI.

59:57We would probably learn from that. Yes. I wonder if there's a YouTube or Twitch channel that does this now. Yeah. We know who would know, right? Who? Well, let's see if we can get back into the audio version of GPT. Okay, GPT. Rick Rubin and I, we're looking for some channels. They could be on Twitch. They could be on YouTube or maybe TikTok or someplace we don't know about where people are speaking to the large language models. And the point of the channel is so others can watch them doing the speaking to learn from it or be inspired or they just find it to be fun. Are there any such channels you can tell us about?

1:00:37Absolutely. That sounds like a fun quest. and there are indeed some channels out there where people basically let others watch them interact with large language models. It's a mix of learning and entertainment. On YouTube, there are a few established creators who do this regularly. For example, you've got channels like AI Advantage, which I mentioned, where the host often does live demos and breaks down different prompting strategies. You can see him talking to the model and explaining why he's phrasing things a certain way. Over on Twitch, you can occasionally find tech or AI enthusiasts who do live streams where they basically chat with an AI and let the audience chime in with prompt ideas.

1:01:16These streams can be pretty casual, but fun and people do pick up tricks by watching. And on TikTok, you can find shorter clicks where creators demo cool or funny interactions with LLMs. They're not always as in-depth, but you'll get a feel for how people are creatively engaging with AI. So yes, there are definitely some fun and educational options out there. If you and Rick Rubin are curious, Those channels are a great place to start browsing and getting inspired. There we go. That was great. And thank you, GPT. You're very welcome. I'm glad I could help out. If you and Rick ever need more AI-related adventures or anything else, just let me know.

1:01:55Have fun exploring those channels. Pretty good. Pretty good. How different are the different AI models available now? There's so many models, but the main models are different from each other, but in some ways converging. So for what you would call casual functions, the GPT models and Claude from Anthropic typically would be considered the best. And they're on a rough par in terms of intelligence. I would say when I want to ask it about music and the history of music, I definitely prefer GPT. But if you wanted it for therapy or for writing or to sound more naturally inhuman, I would prefer Claude.

1:02:33So it depends. If you want to do something quantitative and computational, Gemini typically would be the best. Most people aren't doing that. Then there's models that are especially good at images, like mid-journey. Not something I do, but those are very impressive. You wouldn't say GPT is the best model for images. I think for history, philosophy, discourse, facts about the world, and economics, the GPTs, GPT-5 Pro in particular, would be the best. Do you think that they're programmed with those things in mind? Is that the reason one is better than the other is the way that it's programmed? Programmed is a tricky word here.

1:03:11What's the right word? It's more like baking a cake than writing a program. And to some extent now, the things write parts of their own programs. Different companies have different emphases, but it also just works out that they try different things. And while that happens to work well for something, and then they keep on going down that path because they found something that's working. So some of it is design. Some of it is accident. I would say Anthropic in particular has targeted the corporate market. So their model, and this doesn't matter for my users, but it's very good at running a consistent workflow that won't just screw up because it runs longer in time.

1:03:53So a lot of companies like that. Individual users are not going to notice that. Individual users. ChatGPT is the best app. It's by far the biggest brand name. It's very good for things like planning trips. Very good for medical and legal advice. That's probably what most people are going to use. Have you experimented with either Perplexity or Cursor or any of the secondary apps? I don't know. In Cursor, that's programming space and you combine it with other things. I use Perplexity every time I'm writing. If you want a citation, Perplexity is the best service for finding links. And I find their interface very appealing, very intuitive, super easy.

1:04:39it just finds for me the links I wanted. It's like a better version of Google where I can just say what I want and it does it. It's not based on search terms per se. Even Google is not anymore for that matter. But perplexity for me is now the better Google. If you were to use ChatGPT through perplexity, would you get a different result than if you use ChatGPT directly? Yes. Now, there's many different ChatGPT models and I will switch the model depending on my query For simple queries, I just have it either on instant or take as much time as you think you need. And it does fine figuring out how much time it needs.

1:05:16But for difficult problems, you put it on full-blown research mode, and it could take up to 10 minutes to answer a question. You get a better, more detailed, more verified, fact-checked answer. But it takes 10 minutes, and often it's just more than you need to know. If you're using ChatGPT through perplexity, it's changed over time, but typically you're getting the simpler versions. Again, fine for most queries, but you're not getting the super high-powered chat GPT. In the past, I remember you saying the AI wants to do the least amount of work to get you the answer. Is that correct? It depends what model you toggle to.

1:05:55So the models have gotten more complex and more varied. So if you toggle to the most complex GPT model, which you have to pay for, to be clear, it often does too much work. For some problems, that's fine, or you want to be sure. You don't mind waiting for seven minutes. Great, you do something else. The instant model does just want to answer your question, and it can be a bit sloppy. And just knowing which model to use is of pretty high value. I don't think it's a hard skill to learn. But people who just turn the thing on and use the first model they see, that can be a big mistake. If you're asking for medical or legal advice, it's definitely a mistake.

1:06:32How would you know which model to choose? So, for instance, my wife left her job earlier this year, so she had to evaluate a whole bunch of contracts, what it meant to sign out of the job and be put on different health care, you know, a payment system. And we fed those documents. You just know something like that. You need the most complex, slowest version possible. You just know it. It does not take much brains. But, you know, you want to look up, oh, what year is Elon Musk born? Just try the instant. It's going to be fine. What is agentic coding? Agents are a thing which I do not yet find useful, but they will do your work for you on the internet.

1:07:13I don't code at all. Anything about coding, I'm not going to have a good answer for you. But you also can use agents to do your coding work for you. People debate how reliable they are at the moment. What's an agent? I think of an agent as a person. Agent's not a person. It's like a person, but it happens to be an AI. Typical use of an agent would be, oh, I want to fly to Los Angeles, go book a ticket for me. And the agent will do that. And you can say, find the best price, make sure it's direct, make sure no red eye, whatever. And the agent does that. Again, the degree of reliability is debated.

1:07:46If you ask the AI itself to do that, it wouldn't do it. You need the AI agent to do that. That's right. And there's an agent function attached to most of the major models. You can toggle it on. I don't yet find it reliable enough. The time I need to check it is typically more time than if I did it myself. And I actually have a whole bunch of human agents. Not everyone does, but I'm quite fortunate. But I think within a year or so, the AI agents will be very good and we'll just be able to use them. So the idea of the agent is if you want to interact in the real world, the agent interacts with the real world on your behalf.

1:08:21Is that right? That's right. And you already use customer service for different things. And some of the time that's an AI agent right now. you might not even know. And for simple tasks, they work fine, getting better rapidly. And I think in less than five years, they'll replace really a lot of customer service functions and jobs. What's obvious to you about AI that nobody's talking about? I don't think there's anything right now about AI that nobody is talking about. What's obvious about it is people are attached to it, but everyone talks about that. In a way, we talk about every aspect of it too much, and we spin ourselves in circles.

1:08:59Some people get obsessed with their AIs in an unhealthy way, like anything else, right? How do you rate the answer you get from your AI versus another source? Well, you could check it using another AI. You can check it using books. At some point, you've got to wonder, well, do I trust the book more or the AI more? And you don't know, but you have an extra source. That's very useful. It's very useful just to ask the AI, please check your work. Are there any mistakes in that? And it will check its work. And typically, if there are mistakes, not always, but it will find them with a pretty high rate.

1:09:33Again, depending which model you're using. And that's super useful if it's something important. Just ask it to check its work and you wait a little more. There's no certainty. I don't think there ever will be. I don't know what there is certainty on. But again, you can know a lot more very quickly and just have a lot more intellectual power at your disposal at the margin for free. Do you think most people want the correct answer or they want an answer? A lot of people in life just want an answer, or sometimes they don't want an answer. They want to be led around in circles and just keep on asking.

1:10:10You see this in a lot of religion, a lot of philosophy. If someone actually told you the meaning of life, we might all just be quite disappointed. And you can ask your AI what's the meaning of life. You get a long answer with a lot of taxonomy. It lists the views of a lot of other people. It's a very thorough, reasonable answer, but it never tells you what's the meaning of life. And maybe that is the correct answer, that there's not a single answer to the meaning of life. But we like to keep on asking. So people want many things from the AIs, which are not traditional answers. They want reassurance.

1:10:42They might want someone to argue with, someone to tell them when they're, you know, full of it. Shall we ask AI the meaning of life and see what it tells us? Sure. Now, this is one of the lower-powered models. Let's call this up. Okay, hello GPT. Could you please tell me in your opinion, what is the meaning of life? Absolutely. Happy to share a thought. So, the meaning of life is one of those big philosophical questions with lots of answers. In my opinion, it's really about what you make of it, pursuing knowledge, forming meaningful relationships, and finding fulfillment in your own way. But of course there's no single answer, just the one that resonates with you, personally.

1:11:24But isn't that too subjectivist an approach. That is what resonates with me. I already knew or already was acting upon. I'm looking to you for additional guidance. If you're trying to tell me, well, what's the single thing I should be more obsessed about? What would your specific answer be? Let me give you a more concrete angle. If I had to pick one thing for you to be especially obsessed with, I'd say be deeply curious about understanding other people and their ideas. I think for someone like you, that lines up well with your interests, and it's a single guiding principle. The more you explore different minds and cultures, the richer your own sense of meaning becomes.

1:12:03And when I'm speaking to you, do you know that I'm Tyler? Is that in your memory for this version of GPT? Yes, I do know that I'm speaking to Tyler. Basically, I remember that from our past conversations and from a bit of context you've given me. So this version of me does have that little memory, and I'm happy to tailor things accordingly. And if you had to summarize what Tyler already thought was the meaning of life, what would you say? Sure. Based on our conversations, I'd say you've often leaned toward the idea that the meaning of life is found in intellectual curiosity, cultural exploration, and really just a relentless pursuit of knowledge and understanding.

1:12:40You seem to find a lot of meaning in connecting with new ideas, people, and experiences. So in a nutshell, I'd say you already see the meaning of life as a kind of ongoing intellectual adventure. And if you had to describe what you think Rick Rubin, the music producer, his sense of the meaning of life might be, what would you say?

1:13:00Oh, absolutely. So if I were to channel a bit of Rick Rubin's philosophy, I'd say he probably sees the meaning of life as being about creative expression and authenticity. He's all about helping artists find their true voice and stripping away the unnecessary noise. So from his perspective, the meaning of life might be about living authentically and creating something that's truly your own. Thank you, GPT. That was fun. It was. But what do you think of those answers? It seemed good. Friendly answers. It didn't seem to go too deep. It scratched the surface. That's right. And that's, you could call it the dumber version.

1:13:38And since it speaks, it gives you shorter answers because it thinks you want to talk with it, which was what we wanted. and if you want to pull out the laptop, you can ask the same question, get this long, complex answer. That will take a few minutes. Do you think if you ask the same question on different days, you get the same answer or different answers? It's always a different answer, but I think a question like the meaning of life, you're always going to get a somewhat vague answer that will be not too committal and talk about human meaning and the individual and something very, you could call it American or Western.

1:14:10I don't think it's ever going to give you like a Buddhist answer. or a Hindu answer. It's just not how the thing is built. What does that mean for someone who's Buddhist or Hindu? How useful will these tools be, considering how Western the programming is? Well, it means you have to ask specifically. So if you ask it, what would a Hindu vision of the meaning of life be? You'll get a quite good answer, but it won't automatically default to that. So that's a kind of soft power victory for one version of American thought. I think this will prove increasingly controversial over time as people in other countries use these services more and more.

1:14:48But if you ask for a different point of view, oh, you know, what does Shiite say about the 12th Imam and so on? It'll tell you. But what you have to ask for, it's a slight nudge, and I think it has a big final impact on the kind of content people will hear, consume, and think about. Because so often we just ask it questions, and we don't always phrase it in the from this point of view, right? Yeah. How else do you use it practically beyond preparing for your podcast, for example? I know you spend a lot of time on it, but it seems like a lot of time you spend just seeing what it can do versus using it for a real purpose.

1:15:26Anytime I'm traveling, I use it to plan much of what I see and do. So my wife and I did this big trip this summer through northern Spain and France. We wanted to see medieval churches, medieval artworks from the 13th century, Romanesque, and I just kept on asking it where to go, what to see, what to look for in the churches. On that topic, I would give it an A+. It was just fantastic. And there's no single book or even five books you can bring with similar knowledge, and that you can keep on asking on anything you want to know more about. And then you're in the church, you take a photo of one part of the church, and you say, explain this part of the altarpiece to me.

1:16:06And it does it. So you can put images in, and it can talk about the images. Absolutely. So when I did a birding trip with my sister in northern Colombia the year before, I would take pictures of plants, of birds, monkeys, and I'd say, what exactly is this? And it tells you, it's incredible. So using it multimodal is important. When the audio function works, which now I think is in most countries, if you don't speak a language, you just say what you want in English, and then you'll say to it, well, please translate this into Italian for me, into whatever language for me, and then it speaks that out to the other person, and then you'll tell it, well, take what the person says, and then in audio, translate it back into English.

1:16:49So you have a universal translator now. It works very well for that. You don't always speak the language, right? It's extremely useful. So if you were doing a trip in Spain and you want to see the art that you want to see before AI, how would you have done that research? Well, books. It costs money. You might have to lug them with you, which is a pain. You're not sure you brought the book that tells you what you need to know when you get there, like you bought a book for Madrid, but you end up in Barcelona. There's always this high error rate on human decisions. And the GPT, it just has read not literally everything, but it will know about any topic.

1:17:28And you just keep on asking and you always have it with you. And there's the iPhone version, the iPad, you know, what you have on your laptop. And it just keeps on working. And I'll ask it, you know, because it knows me. I'll say, well, in this town, where would Tyler Cowen want to eat? And those are excellent answers because I've written a lot on the internet on food and dining and restaurants. So there's a big advantage once the AI gets to know you, which it can through your dialogues with it or things you write. So it knows me very well, and it knows what I want to do, see, and eat. Is there any downside to it knowing you in the answers that you get back?

1:18:06It might make it harder to break from how you used to be, but that's your responsibility anyway. People might become lazier, or they'll defer too much to the AI. is it boosting or lowering the serendipity in your life? And this is, again, a question of the default. If you just ask it, you get one answer. But if you say, inject some additional serendipity into your answer, which I've done sometimes, you get something a bit more random. That's up to you, but it is, in fact, up to you. And a lot of people don't give it the extra nudge, and you could say that's an issue. But it's such a cheap, easy nudge.

1:18:42We need to teach people how to do nudges on your AIs because there's a high return to doing slight nudges. Like, the nudge will work. Tell me more about prompting in general. What have you learned about making better prompts? Well, you need to be quite literal. So if you're asking a question about Trump as president, if you just say, what has happened to the stock market since Trump became president? Most humans would interpret that as the second term starting January 20th of 2025. I did that on the GPT. It was my mistake. it gave me a better answer. It started the stock market measurement with when the first term started, which in a literal Mr.

1:19:21Spock Vulcan kind of way is what I asked for, but it's not what a human would have given me. And I saw that I got not what I wanted, and then I asked again. It was fine, but just realize in advance, spell out exactly what you have in mind. Don't be afraid to specify what kind of answer you want. If you want it to tell you what's wrong with something you wrote, you have to ask for that. Some people would say it's like talking to a highly autistic person. But again, it's not a difficult skill to learn. You just have to keep it in mind. How has AI impacted the stock market? Well, NVIDIA is now the world's most valuable company.

1:20:01It's valued at over$4 trillion, as we're speaking. That's worth more than the whole German stock market, to put that in perspective. I believe it's the most valuable company ever. And NVIDIA does a whole number of things, but the most significant profit center is selling GPUs to the major AI companies. And GPUs are part of what makes AI work. And that's just a sign of the market being very bullish that the demand for GPUs, high-quality ones, will be robust. And NVIDIA cornered that market early on. Originally, they used them for video games. So people thought video games were such a waste, whether or not you agree with that.

1:20:39The device we built to make video games work turned about to be the thing that made artificial intelligence really powerful. So that's an incredible origin story. And NVIDIA just got bigger and bigger as a company. And they used to be dismissed like, oh, they make the stuff for those video games. And now it's$4 trillion. How often is that the case where a company makes something for a particular purpose that people don't take seriously? but then eventually that same thing has great value for another use. Absolutely. Something like computers at first, people don't know. How important will this be?

1:21:13Social networks. Oh, it's nice. The students at Harvard, they can date each other. But obviously, Facebook now, Meta, is also one of the most valuable companies. I think they're in at about$2 trillion or$1.8 trillion right now. Obviously, major force of impact. You know, it's not going to be the case of the food company. new innovation. It's hard to know at first how important it will be. So if the stock market is impacted by AI in such a strong way based on the success of a couple of companies, does that give an unclear situation of the market? As we're recording at the end of August 2025, the market has been super bullish on AI-affiliated entities and kind of meh and flat on other companies.

1:22:01They're only up a little. The AI affiliated entities are up a lot. These things can change rapidly. The last two days have actually been bad for tech. But overall, that's the market voicing its opinion that a lot of the future growth of our economy will come through AI. What tend to be the things that really impact the market in one direction or another? Some of it is just arbitrary sentiment. Some of it is news items. There is actual real news. So when Trump makes different tariff announcements... Real news, things that are written in the news or things that are true? It can be both. So if Trump were to put a tariff on a country, that would affect the values of all the companies that trade with that country.

1:22:42And the market reflects that information very quickly. It's an estimate. But they're tracking something real. Some companies become more or less profitable, and the market guesses at that very quickly. A lot of times there's just articles, well, you know, we think Trump will be more aggressive, less aggressive on tariffs. The story may not be true, or we think something will happen with Jay Powell and the Fed. The story may not be true, but people chase after common signals or markers of sentiment, knowing they may not be true, but other people focus on the same things. And so you might make some short-term bets on what other people think about the beauty contest that is that piece of news, where you're just trying to guess what do other people think is good or important.

1:23:24And that's more arbitrary and subjective. And both matter. Is that really the nature of the market? It's really just what people think as opposed to what's actually going on? Well, it's both. Longer term, it's more tightly linked to what's actually going on. So America has by far the most valuable stock market where the richest country with the best companies, right? That's due to real things, not just people got drunk feeling good about America. But, you know, minute to minute, hour to hour, day to day, week to week, it's very hard to tie a lot of the share price movements to actual fundamentals.

1:24:01Maybe the market is seeing something the news doesn't, you don't know, but a lot of times, Fisher Black called it noise, Keynes called it a beauty contest, And it's subjective. Can you tell me about tariffs? Explain tariffs to me and the pros and cons of tariffs. That's a very complicated question. Let's just do tariffs in 2025 because they have a long history. The tariff rate with the United States and its trading partners has been falling since the late 1940s. There were a series of international agreements. First, it was called GATT. Then it was the World Trade Organization. Then a lot of bilateral trade deals.

1:24:39where countries agreed to lower their tariffs so they could trade with each other more readily. When Trump came into office in 2016, there was a lot of talk about tariffs. He did some, but he didn't actually do that much. COVID was a distraction. Congress had other views, other reasons. Now, in the new second term, the Trump people did a lot more preparation, and Trump had this event. He called it Liberation Day, and he announced tariffs will be very high on a lot of America's trading partners. When the stock market heard that news, the stock market panicked. It fell really quite a bit. The market felt that America has a lot of successful exporting companies and that if tariff rates are higher, that means less trade in both directions.

1:25:25And also American manufacturers like Caterpillar or Boeing, they would have to pay more for their inputs from other countries because those are imports, they would be tariffed. So those companies that export and have to buy inputs from abroad, they would be less profitable. So prices fell quite a bit. Then Trump talked back or walked back some mix of those two things, the plans for tariffs. And since Liberation Day, there's been a lack of clarity as to what the tariffs will be. Some people in the market think the tariffs will settle into a 10 to 15 percent level on most countries except China, where they'll be much higher.

1:26:02There's a national security argument for China we haven't talked about. Other people think that the level of ongoing uncertainty about the tariff rates is very bad for the market. That will keep stock prices lower. There may be truth to both of those views. I would just say there's high uncertainty and confusion as of late August. Economists typically do not favor tariffs. But I think there's a serious question one has to ask. Our government is, what, at least$37 trillion in debt. Maybe it's approaching$38 trillion. You need to pay off those debts. You may not think tariffs are the best way to do that.

1:26:38That's a perfectly fair opinion to hold. But you need to specify what else we're going to do and then compare that to the tariffs, and all the alternatives will look ugly. You can raise other taxes. You can cut back on people's benefits. So I think the discussion has been a bit unfair to Trump in that people just attack the tariffs, and all other things equal. The tariffs are bad for most Americans, But at the same time, we have to do something about the pending deficit and debt crisis. Now, Trump himself, he can be his own worst enemy on this issue because he'll say, oh, the tariffs, we're going to take in all this money and then we're going to send a rebate to people.

1:27:15But if you send in the rebate, you've done nothing to pay off the debt. Or sometimes Trump will say, you know, we're not really going to have high tariffs. The high tariffs, they're a threat to get everyone to lower their rates to be very low. And we want to get all the rates to be very low. And then in that scenario, it's also not a revenue device. So are the tariffs for revenue? Are they for rebuilding manufacturing? Are they for national security? Are they to actually be a bargaining chip to get to true free trade? Are they to send people this dividend? Like all these five things or more going on at once.

1:27:46Everyone's confused. The market is confused. So there's not a simple explanation of what tariffs are going to do because we still don't know. Now, in the current moment where things are, is the market in pain because of the tariffs? I would say yes, the market would be considerably higher if not for the tariffs. But that's a subjective judgment, you know. I would say tariffs have hurt the stock market less than most economists predicted, and economists should approach that with a lot of humility. Tell me a little bit about the history of tariffs. Well, if you go to earlier American history, most of our government is funded through tariffs.

1:28:26So obviously, you and I, we pay income tax. We're extremely familiar with that in a quite painful way. But until the early 20th century and a constitutional amendment, there was no income tax. So American government was much smaller. Let's say it would be about 5 % of the economy as opposed to about 40 % of the economy. At 5 % of the economy and you have no income tax, well, how do you raise money? So they would levy tariffs on goods coming into American ports. Of course, there was smuggling, evasion, bribery, but it raised enough money to be in many periods, you know, well over half of the revenue for the U.S.

1:29:03government. So tariffs back then, whatever you thought of their effects, you simply needed something like that to pay for government. And that's a longstanding phenomenon in history, not just for the United States, but the U.S., because it's separated by these oceans, has only a limited number of ports. We were able to enforce that well enough that it more or less worked for the size of government we had in the 19th century. Tell me a little bit more about tariffs around the world between countries. Well, so many countries have long, in previous times, open borders with each other. So tariffs in those cases do not necessarily work well.

1:29:39People would just find a different way to get goods through the border. But you go back to medieval times or even later, You know, Europe, you have goods transported through rivers because river transport is much cheaper than land transport. So you'd have every medieval prince or duke set up, they'd call it a toll, but it's the same as a tariff on the river. And the ship going down the river would have to pay the toll. And the river's not that big. It's very hard to evade that kind of fee. So it was like a tariff. It paid for a lot of early government. But government in these earlier times, it's often not about public goods or helping the peasants or the citizenry.

1:30:15but it's just about more resources for the nobles, for the government, for the papacy, whatever. So tariffs historically in the longer time view are just a way of extracting revenue and oppressing people. So that's part of the history too. What is VAT tax? From the point of view of someone paying it, it feels like a national sales tax. It actually works in a somewhat different way. But in essence, what it does is it makes goods and services more expensive. So the European Union, a lot of government is funded through VAT taxes. It's different from a national sales tax because it's levied on every stage of production.

1:30:52So if a good is produced across seven different stages, there's inputs. You sell to the next company. They add some value. They sell to the next company. Say there's seven different stages. You could say the sales tax is broken up across the seven different stages. That makes it easier to enforce because you have companies reporting on each other. Give me an example of that. Give me the seven stages. I don't really understand that. Let's say you have a farmer and the farmer grows food, but they just produce the raw stuff, right? It's sent somewhere to be processed. It's then sent somewhere to be packaged.

1:31:26It's then sent somewhere to be transported. And it's then sent to a retail outlet to be sold. That'd be like five stages. Okay. So you look at the difference in value across stages and you tax the producer on that difference in value, and they pay some percentage of that. And in essence, you're raising the costs a bit at every stage. But at no one stage, let's say the VAT is at 15%, at no one stage did the costs go up a full 15%. Say there's five stages, they're each like, you know, paying the 3%, it adds up to 15%. Prices end up 15 % higher. Your stage, you're paying 3%, your incentive to evade is much less.

1:32:05Also, the suppliers and demanders on each side of you, they're sending in their own tax reports and for everyone to collude. You know, if you try to cheat or lie in such a way to make your burden lower, it makes the burden higher for someone on either side of you and they'll turn you in. So it's somewhat self-enforcing. It's quite ingenious system. I would say it's too efficient. Europe is overtaxed. Depends on the country, of course, but you have high income tax rates in most of the EU and you then have value-added tax, which makes goods and services cost much more. So in a country such as France, the tax burden I think right now is very close to 60%.

1:32:45That's quite high. Again, it will vary with the country. Switzerland is a much lower tax country. It's also wealthier. I think a better place to live for that reason. So in Europe, from my point of view, there's too much government, too much government regulation, not enough private initiative, not high enough private awards for private initiative, not enough innovation. and the continent is slowly dying economically, and it's made these other decisions, like with immigrants, whatever you think of them, you can only make that work if you have a decent rate of economic growth. And if you have too much government, in my opinion, you're not going to get a high enough rate of economic growth.

1:33:23So Europe has these massive multiple problems. No one of them is explosive, but it's this slow corrosion of Europe's ability to be dynamic and relevant. And you see that all the time. So the United States, we could solve our fiscal problems, at least in the short run, just by slapping on a vat. We all know it works. Plenty of other countries have done it. It's sort of the easy solution. Like I said before, I hate that solution. It's in a sense too easy, too efficient. You get this big pot of money. My fear also is that America will just up the government spending all the more, so we'd recreate a new fiscal crisis in the future, but at higher levels of government everything.

1:34:02So I don't want the VAT, but if you just ask a lot of economists, what's the efficient way to solve this physical problem? They'll all tell you a VAT, and they're not wrong. I think they're wrong in some deep sense, but that it's the simple way to do it is obviously correct. But it seems more like a Band-Aid to you than a real solution? More like a Band-Aid, and it makes us much more like Europe. And I think the United States needs to stay distinct from Europe, more innovative, more dynamic. They free ride on us. We pay for a lot of their defense. you know all these arguments, but they're basically true, right?

1:34:33But the question is, what else can we do? So we could cut benefits by a lot. Rather than argue that, I'll just predict we're not going to cut benefits by that much because voters hate it. And a lot of benefits go to older voters and older people are more likely to vote. And we're getting more and more old voters as the country ages. And then the other solution, which is what we did after World War II, and this is ugly, but it's to do a mix of what is called financial repression and inflation. So we will inflate away a lot of the debt and put strictures in place that make it hard for people to just take their money out of the banking system or other financial institutions.

1:35:16So we'll erode away the value of the debt using inflation, and that will make people poorer. My intuition, I think about this a lot these days, I actually prefer that to the VAT. Like, it's terrible. It makes people worse off. So does the VAT. But precisely because people hate it so much, it's temporary rather than permanent. Whereas you have the VAT, you become addicted to the revenue. It just goes on forever. I probably would prefer inflation plus financial repression. We, by mistake, did some of that during COVID. We had 8.9 % inflation. In real terms, it lowered our debt quite a bit. It wasn't anyone's call.

1:35:52we just made a big, stupid mistake. But it kind of worked. For a few years, it gave us some breathing room. So I think we're going to do it again. The markets seem to be expecting higher inflation from the United States. Trump signals he wants lower interest rates, which is often connected with higher rates of price inflation. Man, it is ugly and painful. I would never say I'm for it. I'm against it. But at the end of the day, there's$38 trillion in debt. And at some point we need some answers. Is there no solution that's not painful? There is a solution that's not painful. I wouldn't bet on it, but some people think the dividend from AI, which we talked about before, will be so high, productivity will go up, and we'll be able to afford the whole thing.

1:36:36So there's different guesstimates here, but if AI could raise our growth rates by a full percentage point, we'd definitely be in the clear, and it doesn't even have to do that much for us. So think Think of the historical U.S. growth rate per capita, 2.2 % on average. If AI could get us to like 2.8, my guess is we're in the clear. If it gets us to three, we're in the clear. I hate betting on that. I mean, what you're risking. Without betting on it, what do you think the odds are? What's your guess? 30%. Like, it's not so unlikely. I'm making up that number. It's not science. But it's not 1%. You don't want to bet the farm on it.

1:37:16Are there things that could be done that would move it from 30 % to 50 % or 60 %? So far, the Trump administration has been very liberal and lax in regulating AI. And I think that's the right policy. The Biden people were not, or became the Harris people, were not going to be that way. So that's one of the things I've been most positive about the Trump administration. But at the same time, how good it's going to get is probably already locked into the technology. and we're just going to see and no one knows. So the things we could do to help it, we're mostly doing. The thing we need is to make it easier to build local data centers in the United States.

1:37:58So an open AI or Meta or Elon needs more power to run a better model, that they get the more power. We'll see how that goes, but it's a messy state and local government kind of thing. My guess is with 50 states, we'll find some places that will do it. But again, that's a bit of a crapshoot as well. And there are also plans to use United Arab Emirates and possibly Saudi Arabia to build out power sources, which they can do through political means that are not available in the U.S. because it's not democratic with checks and balances in the same way. So whether or not we use their power sources, I think the chance is excellent that places like Europe, India, a lot of the world will buy AI services where the power is generated in the Gulf.

1:38:44The Gulf will become a much more important part of the world. They'll have a second thing that is not just oil or gas. Now, the power producing the AI will, in part, come from oil and gas, of course, but also it will come from solar. It will come from nuclear. UAE has plans to just build a lot of nuclear power plants out there in the desert. Just put them there. No one can tell the government no. Citizens cry, NIMBY, we don't want it. It's like they're just going to do it. So that will matter a lot for the world, whether or not it matters for the U.S. Is the impact from those power sources global or local?

1:39:23The negative impact? I don't know if it's fully global. So say the UAE proceeds with current planning. So they have incredible solar power, because where it is, the sun. They can build incredible nuclear, which they say they're going to do. I believe them. And of course, they have plenty of oil. So they're this triple threat energy superpower. So they're well positioned geographically to sell to Africa, to Europe, to India. I don't know how much further that will go. You wouldn't say that's global. I don't think they're going to sell to Uruguay. But that's a lot, right? It's a very significant part of the world.

1:40:01And I think over time it will be less and less oil and more and more solar and nuclear. So if you think there's this global impact from climate change, over time, we're going to master that. I'm worried in the short run, but longer run, the other power sources are going to work. We know nuclear works. Nuclear fusion might work. If we're offshoring it because there's some feeling that it shouldn't happen in the United States, if it's happening somewhere else in the world, isn't that the same negative to the U.S.? Does it matter where it's happening? Oh, it does. I strongly prefer it happens in the U.S.

1:40:36But say we manage it in the U.S., I'm happy for it to also happen there. Yeah. Right? You know, let them do their thing. And these other countries are going to buy it from somewhere else. It could be from worse places. You don't want it to be, say, from China. And UAE does depend on the United States for its defense. So you can debate how loyal they are. But as places go, you could do much worse. But the notion that it's possible— Why can't it be in the U.S.? You try to build a plant. It can take 10 years to get all the permissions, environmental review. You may not succeed in doing it. There's different workarounds.

1:41:15The Trump people talk about loosening those rules. But again, most of it is state and local. I think the Trump people are good on those issues. You need states to cooperate. In essence, what you do is you bring states deals and you say, this will get you a lot of jobs and revenue. And with 50 states, I'm cautiously optimistic. Places like Louisiana, Tennessee, Texas, we're seeing real progress. But still, there's so much regulation. It can be city, county, state level, federal, at multiple levels, multiple agencies. If it's nuclear power, the NRC is involved. When was the last time they approved anything decent?

1:41:52The EPA is involved, right? Citizens groups. It's a nightmare. Now, that's part of our democracy. It has some other benefits, but I think it's gone way too far. And I hope AI and also nuclear power give us some reason to clear away some of these obstacles and push things through some more. This is sometimes called the abundance movement. Is regulation the biggest problem? In the United States it is. Yes. And if you're regulating something because you think it's toxic, but then it happens in China and you have the same toxicity being produced somewhere else. Probably worse. Maybe worse. Right. I'm confused by the strategy.

1:42:34Say you're a homeowner. You know, you always want it somewhere else. Like, oh, I want the U.S. to do it. You know, let them put it in Alabama, not in North Carolina. So there's this NIMBY, not in my backyard, we'll do it in your backyard. It's a collective action dilemma. And you have these fights. But a lot of state governments want to see it happen, I think, or localities, to get the jobs and the money. But again, it's also a question of speed. So say a state can make it happen, but it takes eight years, and the UAE comes along and says, we can do it in four years. Why would that be the case?

1:43:08How could they do it in four years? They're an autocracy, right? And it's a small place. They don't really even have federalism the way the United States does. They have different governments within the UAE, but within, say, Abu Dhabi, they can just decide to do something and do it. They're federal at the level of there's different governments within UAE. But they even are competing to get it done more quickly, in one way of thinking about this. So they just announced it's going to be done. China, not the same system, but they have similar powers in the sense there are no democratic checks on what the government does.

1:43:45And China can get train lines built more quickly than we can. That's obvious. There's also downsides to that. But when it comes to AI, I think it's mostly upsides. But imagine if it gets done first in UAE or China, and that the world's smartest entities are not in America. This I find highly disturbing. I don't know what the actual concrete problem is. It just seems to me when you give up on that, you've relinquished your status as world leader, and that will be followed by other mistakes as well. And imagine the Chinese have a military powered by smarter, say, maneuvering. Their drones maneuver much better, faster, more coordinated than ours because they have better AI because they got the power built more quickly.

1:44:31I don't think they're going to take over California, but globally in any conflict, we're toast. A few months ago, there was a big story about the first Chinese AI that came online. And the reason it seemed to be such a big story is that it was built so cheaply compared to what was built in the U.S., and it was competitive. DeepSeek, that's right. Tell me about DeepSeek. DeepSeek came out of a Chinese hedge fund, and it was done, according to them, by a remarkably small number of people. Some estimates have said 150 people built this incredible system. That fact alone is mind-blowing. Now, they had to be super smart and highly trained and well-paid.

1:45:12but the notion that 150 people can build something amazing. Old companies were not that way. Like General Motors, Ford, that was amazing in its time. How many people did they need? Well, a lot more than 150. So that should give us pause right there, but in part because we did not allow our best chips from NVIDIA to be sold to China for very long. There was like a four-month period where the Chinese could buy them. But we placed export restrictions on those chips. That was part of the Biden legacy. So China, because they had inferior chips, had to innovate in other ways. And they made the thing in a cheaper way that required less underlying force.

1:45:52So that was a big advance. Now, it's complicated what's happened with DeepSeq. They haven't had a really great follow-up. And we're not sure why. It's not transparent to outsiders. So again, you can't believe everything you read. But the story seems to be the Chinese went to the DeepSeq researchers and took away their passports and said, you're too valuable to leave China. Which, in a sense, they are. Like, if you work for DeepSeek, you're being paid a lot by Chinese standards, but you have companies like Meta offering pay packages. We're not sure how much. Some even close to a billion dollars, but like at least several hundred million dollars at the very least.

1:46:31So the DeepSeek people might leave China. There's another Chinese AI company, Manus, which reorganized in Singapore outside of Chinese control. China, of course, upset about that. But now you know, well, you do something great, they're going to take your passport away. Like, what's your incentive? You know, it's not a good place to be. So internally, I don't know anyone who knows what's going on at DeepSeek. It could just be it's technical trouble, they'll come out with something amazing, but it could just be their incentives are screwed up. And they're under probably a lot of surveillance, worries about spies.

1:47:08I don't know. It's a complicated situation. Based on what DeepSeq was able to accomplish at the time that they did, did U.S. companies learn anything from DeepSeq? We learned a lot, and we learned we're going to have a lot of competition no matter what. We learned particular things from DeepSeq. So GPT-5 is a much more efficient model than its immediate precursors. And this is proprietary information. I don't know, but I suspect in part People borrowed the improvements from things like DeepSeek to make our own systems better. But the most important thing we learned was something can come out of nowhere.

1:47:44So the Chinese Communist Party, I believe, was more shocked than anyone that 150 people on their own built a thing that is smarter than the Chinese Communist Party. And to them, that's a threat. Now, our government, Trump talks about AI. We were watching his press conference last night. He brought it up four or five times. No one made him talk about it. But he said, well, it's this great promise. Of course, there's some problems. I felt I could tell just from the language he used that it came up recurringly. It's something he really thinks about, and he thinks it's very important. But he's not doing things to stop it.

1:48:21Chinese Communist Party, this is speculation on my part, I think they have backed it, but they're terrified by it. Every day they wake up and thinking, should we, you know, pull the kill switch on this? Now, this goes back to one of our earlier points. So if these AIs are broadly American in their orientation, and DeepSeek is, if you chat with it, it's not that different how it thinks, how it answers than the U.S.-based systems, because in part, it's building on them in different ways. So it's true that it's censored. It won't tell you about Tiananmen Square or whatever, Taiwan. But the way it thinks is pretty American.

1:49:00and that the smartest entities in China will think like Americans. I think the CCP is smart enough to be afraid of that too. It's a very slow development. Nothing happens for years. But over time, you're turning your everything over to American ways of thinking. What's the new development that's passed large language models? I believe it's the ex-CEO of Twitter who Elon fired, who has a new company now using a different method of AI that's not a large language model? We haven't seen those models yet. I hear speculations. I guess I'm skeptical. But, you know, there will be future innovations at some point, and they're not all going to be large language models.

1:49:49But I feel if it were more of a thing, I would have heard more than I have. But look, DeepSeek was a shock, so there'll be other shocks. But at any given week, I would say bet against the shock at this point. What could it be that's not a large language model that would be an innovation? There's this other approach called neurosymbolic AI, and I suspect in the long run we'll integrate large language models with neurosymbolic approaches. What does that mean, neurosymbolic? GPT can explain this better than I can. Can I turn over the floor here? Please. Let's see what it has to tell us. Okay, so neurosymbolic artificial intelligence.

1:50:28What's a good definition of that that someone could understand on a non-technical level and include a contrast with large language models? Sure. So think of neurosymbolic AI as a blend of two worlds. On one hand, you have the neuro part, which is like the neural networks and pattern recognition you see in large language models. On the other hand, you've got the symbolic part, which is more about using explicit rules and logical reasoning, almost like how you'd use symbols and facts to solve a puzzle. So neurosymbolic AI tries to combine these, letting the system both learn from data and also reason in a more structured way.

1:51:04By contrast, large language models like me mostly just learn patterns from huge amounts of text. We're great at generating language, but we don't have that built-in logical reasoning with symbols. Neurosymbolic AI is basically trying to give AI the best of both worlds. But since large language models in recent times have so much outraced neurosymbolic approaches, should we be at all bullish about the neurosymbolic approaches? Haven't they been failing us and will really do better by just applying more large language model techniques? It's a really fair question, and it's true that large language models have been on a kind of explosive run lately.

1:51:42They've outpaced a lot of other approaches in terms of capabilities and just sheer practicality. But that doesn't necessarily mean neurosymbolic approaches are doomed or that they won't have a place. In fact, a lot of researchers and thinkers are still pretty bullish on the idea that combining the two approaches could lead to more robust and reliable systems, especially in areas where you need more structured reasoning or where just pattern matching alone might not be enough. So while the spotlight is definitely on large language models right now, there's still a case to be made that neurosymbolic methods might give us some extra tools and a bit more flexibility down the road.

1:52:16Thank you very much. You're very welcome. Anytime you want to dive into more AI topics or anything else, just let me know. You talked about the DeepSeek engineers having their passports taken away. There are stories now about AI engineers being recruited from competing companies. Meta's trying to hire them. Grok is taking people from OpenAI. Tell me a little bit about this world of competition for these engineers, and what are some of the reasons that the engineers choose to leave one company and go to another? Some people are motivated mainly by money, others by work conditions, others want to do the most important thing, or they dislike the culture at a particular AI company.

1:53:02If you look at the private market valuations for OpenAI, I think the next round is expected to hit$500 billion, which would make it the most valuable private company ever. I believe. Anthropic has tripled in value since March 2025, and this is only the end of August, 3x boost in basically half a year. That's pretty incredible. So just the financial stakes are very high here, and NVIDIA actually does AI too, not just chips, and they're worth$4 trillion. Meta and Google are more complicated because they earn money a lot of other ways, but they have AI products and are making a big push in this market.

1:53:43But the number of people who know enough to do it or smart enough to do it, who are willing to give up most of their private lives to do it, it's limited. It's extremely limited. And there's also national security considerations. So a lot of them come from China. Some very high percentage of AI researchers are Chinese nationals or were Chinese nationals. They get hired, but still, the Pentagon, the CIA get worried about this. And the more you worry about it, the more that is going to restrict future labor supply. And now the Chinese companies have a demand for labor as well because they have good products.

1:54:24So at the end of the day, I recall it was not that long ago, if you were a very good AI researcher, you'd earn$5 to$10 million a year, and everyone was super impressed by that number. And now that's sort of peanuts. And again, what's rumor? What's truth? It's not completely clear. But Meta in particular made a big hiring push. And there were rumors that I'm pretty sure were at least minimally true of people being offered$300,$400 million, possibly some offers approaching$1 billion. For an employee. For an employee. But keep in mind, you know, so DeepSeek was 150 people. OpenAI, when it made its main breakthrough with GPT-4, the number would fluctuate week to week.

1:55:08But think of the core team as being about a dozen people when they were doing their most important work. One dozen? One dozen. Twelve people? Twelve people. There's different ways you can interpret that. Some people help them. There's like A-team, B-team, and so on. But very small numbers. And it seems now, even very lately, Meta has decided they have too many people in the way and that you'll get better results by having a small kind of... Elite team. Elite team holding the whole project in their heads. And this is very different from how almost anything else is produced. It's like a music album in the 1960s.

1:55:43Like the Beatles make Sgt. Pepper. There's four of them, but like five, George Martin. Paul and George Martin doing a disproportionate share of the putting it all together work. And putting 40 people on that project makes it much, much worse, right? Yes, yes. You don't even have anything. No. So this is a bit like that. So if you're one of the key core people, you're worth an incredible amount because you're helping these companies get to valuations of many, many billions or trillions even. And to pay someone a few hundred million, in a way it's, I wouldn't say it's peanuts, but also think about it from the point of view.

1:56:23Let's say you're Mark Zuckerberg and you have a controlling share in Meta. I don't know what Mark's personal fortune is. You can't really trust what Forbes tells you. But let's say it's at least$150 billion, which is not unreasonable. It could be more. Who knows, right? And the company's worth in the neighborhood of$2 trillion. Whatever Mark wants to buy, he can buy. So if he spends some of what is, in a sense, a lot of it is his money, hiring top people so he wins the race. If I were him, I'd spend the money on that. Yeah, because if he wins the race, it's worth so much more than what he's spent.

1:57:02But it is worth a lot more. But even just to win the race is worth that much. I hear that, yes, the ego of winning the race is worth something. But the actual benefit really is if the multiples are what you're saying. Whether that gain is captured by Metta or some other winner, we don't know. Is there a single winner? one thing about all this hiring back and forth is new ideas spread across the labs quite quickly, as you would expect. Yes, everyone has signed an NDA, but at the end of the day, you know a bunch of techniques that won't work. And I don't care what your NDA says, it's not illegal to simply refuse putting time into things you know don't work.

1:57:46So ideas spread, practices spread. I don't know that they'll ever be like a winner. But again, if I have that level of wealth, I want to win. Some of it's ego, but a lot of it is you want to help the world or you just want to win or a lot of different reasons. Some of it is it could be worth a lot of money. So maybe some of these people are still underpaid. How many winners do you think there'll be in this race? More than one. So I would not bet on a single winner. Early on, people thought that would happened. Opinion has moved away from that view. I'd guess three to four. So right now I would call Google, OpenAI, and Anthropic co-winners.

1:58:33You can have your preferences, and they're better at different things, but they're more or less tied for first. So right now you have three winners. You have Grok and Meta nipping at their heels, trying things, uncertain how well all that goes. You could end up with five winners. Something from China might work. I don't think you're going to have 10 winners. Do you think there'll be new entrants in the race? Right now, it's hard. To get GPUs costs a lot of money. To get this talent costs a lot of money. New entrants from China, maybe. But right now, I think what you see is what you get for a while.

1:59:10And there's five, plus China. Nothing in Europe. Nothing. And keep in mind, the Chinese models, they so borrow from what the Americans have done, or you could say what the Chinese in America have done. They're not that independent as models. And there's debates how much is original, how much some people even say was taken or trained upon. I'm not sure, but they're not really quite original models. They're original modifications to things American companies did. Do you remember when the EU was formed? Well, there's different versions of the EU. If you go back to the EC, it's quite old, but the EU is a strong regulatory body.

1:59:53I would date to the 1990s. And what was the original idea of the EU when it started in the 1990s? Well, the original idea of the European community was free trade and investment and over time migration. And it was a very sound idea with a limited number of members, which had more or less truly common interests, like France and Germany after the war, and you bring in Belgium, the Netherlands, Luxembourg. They're next to each other, common worldviews. It worked very well. It enabled rebuilding. There was this longer-term vision that might help with defense. But it grows and it grows. The current EU has 27 members, some of them very poor, some of them sympathetic to Putin, some of them hate Putin.

2:00:39Whatever you think the right view is, that's a problem. You have a unanimity rule for key decisions with 27 members. That doesn't work. Nations in the Balkans just think about the world very differently than, say, the Netherlands. Obviously, the UK left. They regulate far too many things. The European Commission is increased in power, European Parliament. They just view themselves like America creates, China builds, and we regulate. And if that's your self-image, maybe I'm too American in this perspective, but I think that's kind of sick, actually. You don't have a future doing that. But that's the status quo.

2:01:23So I think the EU is a mess. I still think it does a lot of nations a great amount of good. Someone like Poland. Poland now is as rich as much of Western Europe or it will be in two or three years. And it's a stable country. It's, in broad terms, a free country. would not have done that without the EU. Croatia, Slovenia has done very well, but it's out of control, and on immigration, it's been a disaster. I think immigration policies should be set much more at the national level, whatever you think they should be, and countries are splitting away and defying EU law in different ways and doing this.

2:02:02I think that's healthy. I do think you should have free immigration within the core wealthy, similar EU nations, which would include Italy, Spain, Portugal, Germany, France, Belgium, Netherlands, Luxembourg, a bit more to the north. And after that, I would say it's unclear. Regulation doesn't produce anything, is that correct? It stops people from producing. So there's an EU AI Act which is going into effect in stages. it's being put into place now. It seems odd that it would come together with the idea of regulating. I don't know why anybody would have chose to join it in the first place if that was its mission, was just a regulation body.

2:02:46That's not all it does. So for the poorer countries to the east, it's valuable protection, it's a valuable set of trade relations, and it stabilizes your government just to be part of this exclusive club. It's very good for just about all those countries. For the wealthier, more powerful countries to the West, it's less clear increasingly what they get out of it. So if you over-regulate AI, like does that really harm Bulgaria? Is Bulgaria going to have its own AI companies of import? Probably not. But like Romania actually might have some smaller suppliers, but it's hard for them. So they want to leave to go to London, want to leave to go to the Bay Area.

2:03:26You get this brain drain, company drain. All the time, people get in touch with me. They want me to help them come to the U.S. to get O-1 visas. And so often, they're from Europe. And I'll ask them, you know, well, isn't Europe great? They say, yes, I love it. I love my country. Then they say, I can't do business here. And they want to leave. It's a huge problem. And it's not the regulation in a particular country. It's the EU problem. There can be both, to be clear. But typically for AI, the EU is the villain here. And even Germany and France at the national level have wanted it to be looser.

2:04:05Not loose enough, I would say. But they've pushed back a bit against the EU, made it a little better. Do you think that the people who read the EU just don't understand it? I think they're suspicious of it. So they value the European way of life. And any change to them is a threat to that. And I don't think they're wrong. AI will change Italy. The internet has changed Italy and many other places. But at some point, you've got to figure out a way to make the changes work for you. You cannot keep all your fingers in the dike forever. And I think we're at that point now. But I think that they are so suspicious of it, it is a sign of their intelligence, not their stupidity.

2:04:46But again, either, say, many parts of Italy just become a museum, or they find a new way to be great in some way that's different from how they were great in the past. Now, some of Europe, I think, should be kept as a museum. But to afford the museum, you need economic growth, you need well-functioning, flexible economies, you need the ability to protect yourself against hostile powers, the ability to stand up to China and many other things. And you need some more wealth for that. You need more top companies. You look at all the countries that are in Europe and you look at what are the biggest companies, when were they founded, typically the answer is something like 1920.

2:05:53TATRAGRAMATIN Tetragrammatin is a podcast. Tetragrammatin is a website. Tetragrammatin is a whole world of knowledge.

2:06:08What may fall within the sphere of Tetragrammatin? Counterculture? Tetragrammatin. Sacred geometry? Tetragrammatin. The avant-garde? Tetragrammatin. Generative art? Tetragrammatin. The Tarot, Tetragrammatin, Out of Print Music, Tetragrammatin, Biodynamics, Tetragrammatin, Graphic Design, Tetragrammatin, Mythology and Magic, Tetragrammatin, Obscure Film, Tetragrammatin, Beach Culture, Tetragrammatin, Esoteric Lectures, Tetragrammatin, Off the Grid Living, Tetragrammatin, Alt Spirituality, Tetragrammatin, The Canon of Fine Objects, Tetragrammatin, Muscle Cars, Tetragrammatin, Ancient wisdom for a new age.

2:06:55Upon entering, experience the artwork of the day. Take a breath and see where you are drawn.