What this covers
This retrospective brings together Tyler Cowen reflecting on themes from his 2023 interviews—a format that moves across media, technology, psychology, education, and economics while maintaining a central thread about how learning and cultural appreciation are being reshaped by artificial intelligence. The archive captures Cowen arguing that LLMs and fan-perspective curation represent not threats to intellectual life but opportunities, provided that authors and thinkers embrace the competition rather than resist it. The work is conversational and episodic; it ranges across specific predictions about technology adoption, historical reinterpretation, and institutional change, without retreating into generality.
Several positions stand out for their contentious framing. Cowen contends that most people do not read the books they purchase, and that LLM-powered summarization and extraction tools should cater to this reality rather than shame it. He argues that the humanities are thriving outside academia—more people reading Shakespeare now than ever before—and that this separation from universities may benefit the fields themselves. He defends sociopaths as underrated, distinguishing them from psychopaths and suggesting cognitive diversity deserves more charitable treatment than pathology frameworks allow. He also reframes the Simon-Ehrlich Bet on resource scarcity, suggesting rising prices signal AI's success, not failure. On media, he predicts that specialized outlets using analytics and written from fan enthusiasm—the model of The Ringer—will outcompete generalist newsrooms in coverage areas like sports. Throughout, the thread is comparative advantage: writers and institutions that identify what they uniquely offer, rather than what machines can do cheaper, will survive.
Cowen argues that the future of learning and media lies in AI-augmented, fan-perspective curation that teaches people the art of appreciation, and that this competition should be embraced rather than feared by authors and intellectuals.
- AI tools (LLMs) are already superior to additional book-reading for filling in contextual gaps during research
- Writing from the perspective of an enthusiastic fan (the GOAT/Ringer model) is what audiences actually want and what academia suppresses
- Most people don't read the books they buy, so AI summarization and tailored reading tools should cater to actual demand
AI summarization and personalized reading filters will become standard tools that enhance rather than replace deep engagement with books.
- Most people do not actually read the books they acquire—they buy them and they sit dormant—and there is a lot of evidence from sources like Kindle that this is how most people read, which justifies catering to that demand with AI tools that summarize or extract from books.
“Mostly they just don’t read them. They buy them, and they sit dormant.”
- Within less than two years, and technologically possibly already now, an LLM trained on a specific reader could excise the parts of a book that reader already knows, enabling much faster, more efficient reading; the constraint is token cost, which will become affordable within a few years.
“I think in less than two years. Now, whether that’s a product on the market that I can subscribe to, I’m less sure of, but technologically speaking, it’s probably possible right now. It’s a question of at what cost?”
- An AI reading filter that's roughly 87 percent accurate is almost certainly a help rather than a hindrance, because one can always set the trade-off between type I and type II error—if it leaves in too much, the reader would have faced that content anyway, so there is no realistic scenario where it isn't a net help.
“You can always set type I versus type II error. It almost certainly has to be a help. I don’t see the scenario where it’s not a help.”
- Cowen released his book GOAT as the first book published in GPT-4, a 100,000-word text that can be read traditionally or interrogated through a GPT-4-based app that summarizes chapters, provides background, and identifies the author's possible mistakes—representing the future of much learning.
“Then GPT-4 came along, and I started thinking, “Well, this should be the first book published in GPT-4.” It’s 100,000 words.”
- The biggest change in Cowen's production function since LLMs is that for guests who haven't written much, or for references in books he doesn't know, he interrogates an LLM for background and context—finding this more useful than reading more books, i.e., taking one book and asking questions about what he doesn't know.
“I find that is more useful than trying to read more books, is taking one book and the things I don’t know, asking questions. That would be the main difference.”
Authors must embrace competition from podcasts, AI tutors, and summarization to discover their true comparative advantage in a transformed media landscape.
- Because many more people will play around with LLMs rather than reading books, simply writing isn't good enough anymore, so Cowen has started doing more personal appearances and talks to diversify out of only producing books.
“What I’ve already started doing is, I do more personal appearances and give more talks because I think simply writing isn’t good enough anymore.”
- Authors should run toward the competition from podcasts, AI tutors, and summarization rather than away from it, because confronting it forces an author to identify their true comparative advantage—knowledge most authors never gain.
“I’m fine with that competition. Most authors are running away from it. I think we should run toward it and embrace it. If you have to figure out, “This is actually my comparative advantage,” it’s fantastic to know that.”
- Comprehensive media outlets like the Washington Post and New York Times will stop covering sports because specialized outlets (The Ringer, ESPN, The Athletic, Substacks, podcasts) that use analytics and write from a fan's perspective decisively outcompete generalist sports sections, leaving generalists to cover their specialty like politics—an outcome that is for the better.
“Places like ESPN orThe Ringerjust way outcompete the sports section of, say,The Postor ordinary newspapers that don’t really do sports. That will disappear — them doing sports — one way or the other. And that’s for the better.”
- The fundamental innovation of GOAT is not the 'greatest of all time' framing per se, but writing from the perspective of an enthusiastic fan—as outlets like The Ringer do—which is what many people actually want, whereas academic incentives push precisely toward the opposite, making subjects seem unexciting.
“I think the fundamental innovation is not GOAT per se, but writing from the perspective of a fan. It’s what a lot of people actually want.”
Teaching appreciation and cultivating enthusiasm through visible modeling is harder and more valuable than providing information alone.
- The whole mission of Conversations with Tyler is to teach people the art of appreciation, which requires the host visibly demonstrating his own appreciation, because there is plenty of information available but the hard problem is motivating people to want to study—solved by modeling enthusiasm.
“The goal is to teach people how to appreciate. How can you do that if they can’t even seeyouappreciating? It’s insane.”
- Rick Rubin is one of the greatest talent judges of all time, and doing music appreciation with him teaches a great deal about talent and the art of appreciation, which both he and Cowen are keen to spread to others.
“He’s one of the greatest talent judges of all time, and you learn a lot about appreciation.”
- There should be more 'intellectual romance or gossip' books about thinkers' love lives—such as accounts of central European thinkers before World War I or the Heidegger-Arendt correspondence—because they bring history to life, help you appreciate, and reveal what may actually have motivated great thinkers.
“There should be more books like that. It brings things to life. It helps you appreciate. It makes it vivid. You get some sense of what might haveactuallymotivated them.”
Understanding peripheral contexts like substance use, personal relationships, and cognitive diversity illuminates historical and contemporary dynamics.
- Asking a person how they expect to face their own death is a uniquely revealing interview question because it is hard to B.S. in response without sounding like an idiot, since few people have a ready answer, so you learn how thoughtful they are and how they respond on the fly.
“It’s hard to B.S. in response to that question without sounding like an idiot. So that makes it a Tyler question. You get a sense of how thoughtful a person is, and how they respond on the fly”
- Sociopaths are underrated: they can be super-smart, high-achieving people, and M.E. Thomas's Confessions of a Sociopath is one of the best statements of how to think more deeply about them—distinct from psychopaths—though they can also be problematic.
“Sociopaths are underrated, I think, is a key point of that one. I’m serious.”
- Cowen interacts with few Protestant intellectuals largely because of where he hangs out (academia, Jewish intellectual circles), but also because there is something systematizing about Catholic philosophy that makes him more likely to engage with it, whereas Protestant theology mostly bores him.
“I think there’s something systematizing about Catholic philosophy, where I’m more likely to end up interacting with those people than I would be for the Protestants. There’s a lot of Protestant theology. When I read it, mainly I’m bored.”
- The roles of alcohol and drugs in war and statecraft are an underappreciated but important lens—Vodka Politics was ahead of its time on Russia, and engaging with the role of alcohol is a fundamental component of understanding Russian dynamics, just as many Nazis were on speed.
“The roles of alcohol and drugs in war, people are starting to pay attention to. That book was an early move in that direction, way ahead of its time.”
- The architecture of Geoffrey Bawa in Sri Lanka, especially the hotels, is one of the great glories of the world that most people haven't heard of, and along with surprisingly good Western baked goods, makes Sri Lanka a highly recommended, safe, and affordable vacation spot.
“The architecture ofGeoffrey Bawa, especially the hotels, is one of the great glories of the world, and most people haven’t heard of it.”
- Cowen is willing to 'bite the bullet' that many conditions considered mental disorders are better understood as cognitive diversity—while specific harmful behaviors are a separate matter, in general we should be more open to viewing them as variation rather than pathology.
“You toss me that bullet, I’ll bite it. I’ll swallow it.”
- In the Simon-Ehrlich resource bet, Ehrlich's prognosis is now, in a funny way, the more optimistic one, because rising resource prices are a sign that AI is working—energy prices will rise a lot as people need to compute and build new projects, which is a good thing—making the Simon (falling prices) side the pessimistic one.
“The funny thing I’ve realized is that Ehrlich’s, in a funny way, is the more optimistic prognosis, that rising resource prices is a sign that AI is working.”
- The film Her will go down as one of the great movies of its time, comparable to Frankenstein, City Lights, or Gone with the Wind, partly because it correctly anticipated seamless voice-based AI interaction and a future aesthetic that updates the 1950s-60s look.
“It will go down as one of the great movies of its time, likeFrankensteinorCity LightsorGone with the Wind. Total clear winner.”
Great thinkers and creators are often underestimated in depth, and sustained intellectual engagement reveals capabilities far exceeding first impressions.
- On rereading Adam Smith, Keynes, and John Stuart Mill for GOAT, Cowen was amazed at how much even smarter they were than he had thought, with Mill receiving the biggest upgrade given his roughly 32-33 volumes that are interesting on almost any random page.
“As I reread them, it justamazedme how muchevensmarter they were than I had thought. That’s the main thing I learned from doing that book.”
- The Vishy Anand episode is perhaps the very best Conversations with Tyler episode on human psychology of all 200, particularly through Anand's account of being a bad loser who self-flagellates in his hotel room after composing himself for the post-match press conference.
“It’s maybe our very best episode on human psychology, of the whole 200.”
- Korean movies for almost a 20-year period are some of the best in the world, with the ones that are known at all being quite good at a success rate of about 90 percent.
“Korean movies for almost a 20-year period are just some of the best in the world. The ones that are known at all are quite good at a success rate of about 90 percent.”
AI will mediate personal media consumption, with systems selecting content for users rather than users making independent choices.
- A podcast should ideally be released once every day because there is something socially optimal about waking up knowing there's a podcast for you, where you only listen to the roughly 11 a year you care about most, but that will be a different 11 for every person.
“Ideally, a podcast, I think, should be once every day. That’s not possible for a number of reasons, but there is something socially optimal about that”
- In the future, listeners will not choose their own podcasts; instead they will ask their AI assistant which new podcasts they want to hear, and the AI will decide.
“No, it’ll be up to the AI. You’ll ask your AI, “Which are the new podcasts that I want to hear?””
- Cowen is most bullish on Meta glasses, expected to be priced below $200, which will be what Google Glass tried to be but now succeed because LLMs are so much better—the notion of wearing glasses that tell you whatever you want to know is here to stay, despite tricky privacy issues.
“The thing I’m most bullish about is theMeta glasses. They’re not out yet. I believe they’ll be priced below $200, and they’ll be what Google Glass was trying to be, except now we have all the tech.”
The humanities are thriving outside academia as diverse global audiences engage with classical works and forms through modern distribution.
- Cowen expects that he will face death suddenly and won't be confronted with much of an issue, and Robin Hanson's empirical finding supports this: in people's final moments they are typically very weak with far less reaction than they anticipate, making death less of a confrontation than imagined.
“Robin Hanson has looked into this a bit, and he tells me, in people’s final moments, they’re typically very, very weak and just have way less reactions, period, than they think they’re going to, and it’s not that much of a question. That was his empirical finding.”
- The humanities are not dying but undergoing a huge revival outside the academy—more people are likely reading Shakespeare than ever (especially counting India and China), classical music thrives on YouTube, and being removed from the academy is probably good for them at current margins, while those who study them demonstrate special and unique talents.
“I don’t think the humanitiesaredying, and podcasts show that. I don’t have data, but I suspect more people are reading Shakespeare than ever before.”
- GOATs are by their nature going to be older figures because recent achievements aren't called GOAT yet and may be surpassed soon, but this does not mean modernity is collapsing—we are generating GOAT-like outputs all the time; we just don't yet know which are the most wonderful in their areas.
“it’s going to be older by its nature. It doesn’t mean modernity is collapsing. If someone did something amazing two years ago, say in gaming, it just won’t be called GOAT yet.”
- Despite a current leader, diversity among LLM providers will persist rather than collapse into a single dominant provider, because—unlike Google's free model—there will be price differences (cheaper open source for places like Kenya), and less-gated models (e.g., Elon Musk's with less reinforcement learning, ruder and more interesting; Character AI playing historical figures).
“Well, there’s a leader now, but I strongly believe diversity will persist. For one thing, there’ll be differences in prices, which is not the case at Google.”
Small-scale intellectual communities and experimental social zones will proliferate because they deliver tangible value to participants.
- Having three to five people read the same piece and discuss it teaches you more than most academic seminars, yet academics don't do it enough because there is somehow no space for it.
“You could have a group of three to five people read the same piece and talk about it... You learn more that way than for most seminars.”
- Pop-up intellectual communities and special economic zones—like Vitalik Buterin's Zuzalu in Montenegro or the Vitalia life-extension experiment in Próspera, Honduras—will become much more common because they are appealing to users (meeting interesting people, interesting locations, low cost), and if it can be done it will happen much more.
“I think there’ll be much more of that. I’m going to visit one of those groups in January in Honduras — Próspera. They’re running a two-month experiment. It’s calledVitalia.”
- It is very difficult to get true geographic diversity in Emergent Ventures winners—especially by where they currently live rather than where they're from—as evidenced by never having received an application or winner from central Tennessee, which Lazarus Lake's success drove home as a real failure and hard problem to overcome.
“it’s very difficult to get true geographic diversity in your winners. Maybe where they’refrom, you can do, but where they live now.”