“Deep Thoughts”: Tyler Cowen on AI and the Future of Work

The trade of the year has been short Meta long Twitter, with Meta stock down 73% year-to-date and Twitter up 35%, demonstrating the fragility of tech monopolies previously thought permanent.

“The trade of the year has been short meta long Twitter. A year to date meta stock is down 73%. Twitter is up 35%”

GPT-3 and later models can write coherent essays, book chapters, and debate speeches indistinguishable from above-average human work, creating an existential threat to academic integrity and knowledge work gatekeeping.

“The problem Tyler is that GPT-3 and some of the most recent iterations can essentially write an essay, uh conceivably uh a book, but I've seen essays, I've seen debate speeches uh that certainly uh pass muster. That is to say, you can't, on reading them, tell that an AI produced them. You could quite easily assume it was an above-average student.”

The Ukraine war has destroyed value at asymmetric rates: Russia's GDP down 3%, Ukraine's down 30%+, with ongoing infrastructure destruction; this represents a catastrophe for Ukraine even if the West eventually wins militarily.

“Russian economy down 3%, Ukrainian economy down 30% and probably more than that now that the critical infrastructure is being knocked out.”

Stanford economist Nick Bloom's research shows that despite theoretical objections, many companies are in fact heading toward 3 days per week in-office schedules, suggesting market-based solutions will emerge for underutilized office real estate, similar to how abandoned big-box retail real estate prices will adjust to its actual value.

“Well, our our colleague Nick Bloom at Stanford is studying this, and actually, uh despite what Neil's saying, that's where we seem to be heading now, that many companies are three days a week.”

Iran is domestically unstable and faces serious internal dissent; it may start a regional war (against Saudi Arabia, Israel) to consolidate power through external conflict, making Iran conflict a near-term geopolitical risk.

“Iran is in such uh a domestic uh difficulties that it in fact needs a war in order to suppress uh the internal dissent. And I think pretty soon we're going to be talking about the new war in the Middle East, which will be between Iran and its adversaries.”

Russia and China are not natural allies but have been forced together by US policy; their partnership is now actively aggressive (in Ukraine, South China Sea) and expanding to include Iran, creating an 'axis of ill will.'

“China and Russia are not natural allies, but we have helped them get together. And that partnership is certainly an aggressive, an expansionist one. That's why about a quarter of Ukraine is still a scene of ongoing warfare, and why China is building an enormous military infrastructure in the South China Sea.”

Journalism, Twitter, Substack, and other forms of generic/short-form writing will be overwhelmingly written by AI bots within years, as the incentive structure makes it trivial for writers to input prompts and let AI fill out content, flooding the internet with AI-generated text.

“Journalism, Twitter, Substack, those are just going to be 99% written by bots. If you suppose you're you know your job is to write a Substack you'll just put in a prompt, let the AI fill it out. It's just it's going to overwhelm the internet.”

Taiwan's strategic position (small island defended across vast distance against geographically proximate adversary) makes it very difficult to defend with any technology; even US military advantage cannot reliably prevent Chinese conquest.

“The major risk would be Taiwan falling which perhaps we cannot prevent with any technology and that spreading to a series of other conflicts around the world because American credibility is lower. Very different from the problems of the British but you could imagine if Taiwan falls or is in danger, Israel and Iran start a war, other hotspots break out.”

The US faces a debt-to-GDP ratio approaching 100%, and this decade interest payments on debt will exceed the defense budget, leaving insufficient fiscal resources to fight a sustained war while maintaining deterrence.

“strikes me that uh yes, as I got two economists captive, I should raise the pro prospects of the classic problem of any uh of any great power, the problem that arises when your debt service costs exceed your defense budget. That's going to happen to the US this decade with a very high probability. And it strikes me that that's one other way in which the British problem uh repeats itself.”

Historical pattern shows dominant empires become wedded to weapon systems that achieved their power, making them unable to innovate; the US is replicating Britain's interwar predicament by clinging to obsolete platforms (aircraft carriers) while rivals develop superior alternatives.

“the problem about being number one, uh, about being the dominant empire or great power is precisely that you get wedded, uh, to the weapon systems of the the generation that brought you to that position. And I see plenty of evidence that the US has that pathology, uh, that was very much a part of Britain's problem uh, between the wars.”

US intertemporal national wealth relative to debt is more favorable than China's; the US has superior demographics (higher fertility, immigration), higher-quality capital, and better institutions, making long-term US portfolio more attractive than Chinese portfolio.

“our intertemporal trajectory of national wealth, not GDP, but national wealth relative to debt, I would much rather hold the US portfolio than the Chinese portfolio. And their total fertility rates are low and plummeting. Getting rid of the one-child policy hasn't helped. Ours is lower than it should be, but it's about what? 1.7? We have a much better history of taking in, absorbing, and attracting excellent immigrants, including yourself.”

Historically in tech, every disruptive technology has proven to be a tool enhancing human abilities rather than replacing them, though the distribution of benefits varies (some people's productivity raised, others lowered), suggesting this pattern will hold for AI despite dystopian narratives.

“everything in in tech so far we there's there's the dystopian nightmare of the tech takes over and then there's the factual reality of of everything since this the the first looms that this is a tool that enhances human abilities although raising and lowering some people's productivity.”

The concept of 'fake' content will change: AI-generated or AI-assisted work is not fake in the same way as falsified data—it is more like a book produced by visiting libraries, amplifying human knowledge through tools.

“Would you say to an author, 'Oh, your book is fake. You wrote it by, you know, visiting plenty of libraries.' So, I think our understanding of fake will change.”

The US is pursuing reckless policies in Eastern Europe, the Middle East, and the Far East that risk replicating the British predicament of 1940, where technological dominance proved insufficient against strategic miscalculation and overcommitment.

“My sense is that we're actually on on a on a course to to replicate the British predicament circa 1940 because we're pursuing I think quite reckless policies in Eastern Europe in the Middle East and in the Far East and and not enough attention is paid to that historical danger.”

Xi Jinping is placing China on a war footing; China will likely attempt to take Taiwan within the next 10 years, potentially causing a global military catastrophe that the world must prepare to adapt to.

“I think Xi Jinping is putting China on much more of a war footing than people realize. I do think China will try to take Taiwan in the next 10 years, and it may be disastrous for the world, but I don't see them as an expansionist power in the same way Nazi Germany was.”

AI military applications favor the West and the United States in the medium term because AI development is concentrated in the Anglosphere, particularly the US and UK, while China's AI development remains opaque and unproven at scale.

“This could be the number one issue in the world right now, but I think at least in the medium run, it's a major advantage for the United States. We don't have a number of troops advantage. If you look at the new AI products that we see, they do come from the West and typically from the US, a bit from Southern England. China is mysterious, put a lot of money into it, but they haven't, you know, shown they've come up with much.”

AI will not achieve sentience or misalign its goals with humanity because it lacks sufficient hooks into the physical world; the danger is not AI rebellion but humans (malevolent actors like future Hitlers) using AI as a weapon system.

“Look, what I don't think artificial intelligence is going to kill or enslave us all. It doesn't have enough hooks into the outside world. Our world is fairly decentralized and I don't think it will become sentient in a way that will misalign its goals with our goals. Could there be an evil Hitler who uses AI for evil purposes in the future? Of course. It's true of any weapon system.”

China's debt problem is worse than the US's; Chinese debt is distributed across local governments, special financing vehicles, and real estate rather than concentrated at the federal level, making it opaque and harder to resolve.

“China has a worse debt problem than than we do, right? It's not mainly at the federal level, it's at the local level, it's through special financing vehicles, it's at the real estate level.”

AI researchers who were considered crazy 3 years ago have been repeatedly proven correct by AI's actual capabilities exceeding expectations; we are entering a series of revolutions in AI impact within the next 5 years.

“I talked to a lot of people in the AI area who I thought were a bit crazy 3 years ago. And over the last 3 years, they have been repeatedly surprised by AI itself, and AI is doing far better than what they had expected. And I think their point of view essentially has been proven to be correct. We're in for a series of revolutions, not starting in the next 6 months, but well within 5 years.”

Stable diffusion and related image-generating programs can produce 'quite interesting, quite vivid visual images' from prompts and have become 'democratized much more quickly than people expected,' alongside rapid GPT-3 democratization.

“There are also image generating programs, stable diffusion is the one that is open to the public, where you can feed it prompts and it will give you often quite interesting, quite vivid visual images. Uh both stable diffusion and related programs and GPT-3 are quite recent creations and they have become democratized much more quickly than people expected.”

Germany was technologically competitive with Britain in the 1930s-40s, whereas China today appears much more bureaucratic and economically stagnant than historical Germany, suggesting the US does not face a technological peer competitor as Britain faced Germany.

“Germany technologically speaking was doing pretty well relative to England. I don't see that we have a rival like that. We have a set of allies that China can't match. China itself appears to be much more bureaucratic than we had been thinking. Look at their zero COVID policy and its persistence. Their economy slowing down, population aging.”

Future intellectual entrepreneurs will succeed not by having original ideas in isolation, but by understanding how to frame novel prompts to AI systems and how computational methods interact with domain problems.

“They're the ones who understand how computational biology and solving the biomedical problem interact. They're the new intellectual entrepreneurs. They will present the correct prompts to the GPT whatever. It'll be better than GPT-3. And they will say, 'Search this area space. I think the combination of like fiscal policy, open economy, and impending election hasn't been adequately explored. Please give me a rapid printout on which combinations of those ideas have not been covered, and you'll do a bit of the writing, and it will do a rest of the writing.'”

TikTok's success relative to Facebook and Twitter stems from the power law: with hundreds of millions of short videos created daily, the probability that some will become viral is extremely high, and entrepreneurs have systematically underestimated this multiplicative effect.

“On TikTok, I think the notion that makes it work is every day, it's a bit like YouTube. You have so many swings at the fence. So, you have hundreds of millions of videos being created in short periods of time, and some of them will take off. And I think entrepreneurs simply not taking the power law seriously led them to both undervalue YouTube.”

India is chronically non-aligned and will remain so despite US efforts to integrate it into a containment coalition against China; India is an obstacle to China but not a reliable ally.

“a lot of our strategic thinking has assumed that we can convert India into an ally, but I think India is chronically non-aligned. It was non-aligned in Cold War I, and I think it's non-aligned in Cold War II.”

The optimal hybrid model is broadly remote workers located globally with regular in-person offsites (3-4 days quarterly/semi-annually) that enable team bonding, informal knowledge transfer, and cultural alignment that Zoom cannot replicate.

“I'm going to throw out the idea that what you actually do is have a broadly remote structure with people living where they want to live, but you have to have quite regular offsites where the tribe gathers uh and the Dunbar number permits all those uh things that you can't do over Zoom. And I think that's more attractive to me than a kind of part-time commuter lifestyle and a part-time office uh lifestyle. By the way, that's what we've been doing at Green Mantle and it's really it's really working quite well.”

China's zero-COVID policy persisted despite obvious economic costs, illustrating bureaucratic inflexibility; this suggests the Chinese regime is less adaptable than Western democracies and less able to capitalize on technological advantages.

“China itself appears to be much more bureaucratic than we had been thinking. Look at their zero COVID policy and its persistence. Their economy slowing down, population aging.”

The metaverse adoption problem stems from a category error: social media's power comes from random, low-friction engagement (checking your phone), whereas the metaverse requires intentional, high-friction setup (putting on a headset), making it fundamentally incompatible with how digital addiction actually works.

“nobody says I think I'll take out my phone and waste an hour on Twitter. No. They they're they do it randomly. You have to take this big thing put it on and say I think I want to waste the afternoon wandering around the metaverse.”

Google's acquisition of YouTube was considered insanely overpriced and a copyright nightmare when it occurred, yet YouTube became 'our greatest educational innovation of all time,' demonstrating how entrepreneurs systematically undervalue power law effects in creative platforms.

“I remember when then Google bought it, and the the purchase price was considered insanely high, and that it was a copyright nightmare and wouldn't go anywhere. It's now our greatest educational innovation of all time, when almost everyone was sour on the original purchase.”

AI-assisted knowledge workers will become analogous to 'Supreme Court Justices with a whole bunch of clerks'—humans provide strategic direction and final judgment while AI/clerical staff handle execution work.

“You'll become like a Supreme Court Justice with a whole bunch of clerks who do a lot of the hard work for you, and you get to put the big picture on it.”

If Taiwan falls, it creates cascade of other conflicts—'Israel and Iran start a war, other hotspots break out'—because American credibility collapses and rivals perceive reduced US willingness/capacity to intervene globally.

“If I think about China, the major risk would be Taiwan falling which perhaps we cannot prevent with any technology and that spreading to a series of other conflicts around the world because American credibility is lower.”

Hype about AI is usually correct about long-term direction but consistently overestimates near-term timing by 40+ years; the mistake is listening to timing claims rather than treating hype as a signal of eventual significance.

“First, the hype is usually correct. It's just 40 years too early in many many areas. So, listen to the hype, but don't listen to the talk about timing.”

Tech monopolies consistently miss emerging platforms by failing to develop new technology in-house; they instead acquire competitors after the fact, as shown by Facebook/Meta missing TikTok, YouTube missing short-form video dominance.

“they they they always do. I mean, the story of of uh, Silicon Valley has been one company hits it big with something and then buys up their new technology. They don't ever seem to be able to develop new technology in-house.”

The major wave of coming business innovation will be in hybrid management models (not just hybrid calling, but hybrid team structures), and companies that master these will gain significant competitive advantage.

“I think a major wave of forthcoming innovation will be these hybrid models. And I I just mean how we do the calls, but how we make hybrid management models work. And the companies that master that, I think will have a significant advantage, but it's something we've barely tried to make work yet.”

We are not good at predicting where machines will be most productive; autonomous vehicles have been 'just around the corner' for 15+ years without materializing, but AI has unexpectedly disrupted white-collar work faster than anticipated.

“There was a big hullabaloo about autonomous or self-driving vehicles 10 years ago. They still might be 10 or 15 or even 20 years away. Uh the thing that's hard to automate, it seems to be, is your gardener. And how much white-collar work can be done by AI, we radically underestimated.”

Facebook employees not using Meta's VR products internally is a significant red flag, because Amazon's successful cloud business emerged from internal tools (S3, EC2) that Amazon developed for its own use and then decided to sell externally, whereas Meta's failure to create an attractive internal use-case suggests fundamental product problems.

“I also noticed that So, I'm reading that Facebook employees don't use Meta. And some of the most successful internal technical developments have been like Amazon develops an internal tool. They developed the cloud as an internal tool, and then they decide to sell it. Now, the fact that Facebook employees don't want to use this thing at all, and it's not an internal tool for, say, hybrid meetings does does seem to be a problem.”

Young wealthy people in the Bay Area exhibiting high rates of depression, undergoing therapy for 10+ years, and experimenting with psychedelics suggests something 'badly wrong' with contemporary life that previous generations did not experience, and this malaise is not shared equally across California's immigrant populations.

“I'm struck by the number of Bay Area and startup people who are maybe 30, doing quite well, been seeing a therapist for 10 years, talking about their latest experiment with psychedelics, which by the way, I do think probably alleviate their depression. But to me, that's a sign there's something badly wrong. That there is something basic about life in earlier eras that we seem to be losing our grip on.”

Tyler Cowen proposes an updated correction to the Snow Crash dystopia: the lower classes in California often have more enjoyable lives than his observation suggests the metaverse-escapism thesis assumes, and the striking phenomenon is unhappy wealthy people being obsessed with escaping to the metaverse, inverting the class-based escape narrative.

“part of my suggested update would be to uh believe that the lower classes in California often have more fun. Uh it's the upper classes who seem obsessed with moving to the metaverse. And just ordinary life including in the less nice parts of California, it's still very very good. So people speak of, you know, East LA is a poor part of the city. Uh it's wonderful to be in and go out to eat and and walk on the sidewalks. So to me, what's striking is how many unhappy well-to-do people California seems to have including the Bay Area.”

The discussion of remote work predominantly features a non-representative sample of jobs; many work sectors (law, sales, HR, government) require in-person presence for fundamental functional reasons, and HR professionals are actively hiring for people who will show up in person rather than take paychecks without appearing.

“But if I speak to my sister who works in human resources, where everyone is supposed to be there, her view is, 'I'm looking to hire people who will show up and not just take the paycheck and never appear.' We are talking about a narrow slice of office jobs.”

Competitive sectors (e.g., fast food, service industries) may experience acceleration of automation and AI adoption because they lack the profit margins to fund periodic offsites and remote-work infrastructure, creating a divergence between high-skill knowledge work (which can go remote) and low-skill service work (which either remains in-person or gets automated).

“I worry more about the competitive sectors where there's not the profit margin that you can pay for these get-togethers. Now, a lot of competitive sectors have just stayed face-to-face all along. If you work in a McDonald's and do fast food, right? You're not doing that from Zoom. So, automation may kick in here. Artificial intelligence may kick in here.”

The problem Ferguson identifies—originality trap from pattern-matching AI—already describes academia as run by humans today, so AI adoption would not create a new problem but continue an existing one.

“That's academia as run by humans today. If anyone should know that, it's you, right? So we're in the trap. We're hoping AI will get us out of it.”

Global distributed teams can achieve global economies of scale by gathering in internationally neutral locations (like Amsterdam) rather than US hubs, solving visa restriction problems for employees in African and other regions where US entry is restricted.

“One of my children works for a tech company and they are they're global. They're a remittance company... they wanted to have they they do everything. Everyone works from home, but they get together occasionally. Most recently, they flew everyone to Amsterdam because for most of the people like the ones working in Africa, they can't get visas to enter the US for a business conference.”

The public sphere will become entirely fake (dominated by AI-generated content optimized by algorithms), forcing serious intellectuals to retreat into small private/luxury markets where they can maintain authenticity and avoid algorithmic toxicity.

“I think what happens in the world you're describing, Tyler, is that the public sphere becomes entirely fake. And and people who want to exchange and exchange ideas as as human beings have to retreat into the private sphere knowing that there's no there's no point participating in a sphere in which algorithms award newness scores to content. I'm just going to vanish altogether and and meet a few select people probably in an Italian town.”

High-quality, original writing by recognized authors (Ferguson, Cochrane, etc.) will become a luxury good commanding premium attention and monetization, while generic writing produced by AI will be commodified and devalued.

“Your your key is high quality original writing. You're not going to be obsolescent. You're you're producing the Gucci. You're producing the Steinway piano.”

Meta's metaverse experiment represents a classic American corporate research lab attempting to drive a bold, decisive technology initiative, which deserves applause even if skepticism about consumer adoption is warranted.

“I think of formerly Facebook, now Meta, as a classic example of the American corporate research lab. And they're spending like crazy trying to get the metaverse to work. I'm sympathetic with the attempt.”

The US has failed to supply adequate drone support to Ukraine despite its procurement position and technology leadership, and Russia is using Iranian drones effectively against Ukrainian infrastructure, suggesting the US cannot even effectively support a proxy war ally, much less conduct direct military operations against near-peer competitors.

“We're not in fact building the kind of defenses, and it's not clear that we're able even to supply them to Ukraine right now where we're fighting a proxy war, a war that's going quite badly since the Russians started to target Ukrainian infrastructure using amongst other things Iranian drones.”

Public intellectual work can function as a 'bat signal' to attract talent, and visibility through blogs, YouTube, Twitter helps identify high-quality people who see the work and want to connect, creating beneficial selection effects.

“we believe we have good means of identifying young talent and advancing it. And I think that will not be disrupted for some while. So, I have a diversified portfolio, and I think of my public intellectual outputs as part of the bad signal that like draws talent to me or to us”

Public intellectuals will evolve into 'impresarios' who curate and direct AI tools and human collaborators rather than producing original work directly, much like how authors adapted to blogging, YouTube, and Twitter.

“I think public intellectuals will be a bit more of an impresario, juggling a bunch of AIs and also human collaborators. And just as I moved into blogging, moved into YouTube, moved into Twitter, in all cases pretty early, uh I'm at this moment thinking through what I will do with AI.”

AI systems trained on past content will enable readers to rate content by 'newness score'—measuring how much of a piece contains genuinely novel ideas versus recycled existing thought.

“You can imagine readers all having their AI programs that'll just scan all the text out there. Like, 'Oh, Ferguson's latest column. How new was that?' It'll have a newness score. And if it's really new, it'll have a high newness score. If it's not, it won't.”

Hybrid meetings (some people in room, some on video) are currently the worst type of meeting, and the metaverse could potentially solve this by creating visual/presence equality between remote and in-person participants.

“Well, first let me say I think hybrid meetings maybe the killer app for the metaverse. So, we're all on Zoom. That works okay. If we're all together, that works okay. If half of us are together and half of us are on Zoom, that works terribly. So, the potential productivity gains from mastering hybrid meetings, I think are very, very high.”

The US can impose constraints on military AI (e.g., limiting the number of pawns it sacrifices) while still achieving military advantage through better human-AI integration, drone swarms, and tactical decision support at squad/platoon level before fully autonomous AI generals emerge.

“you can tell your AI not to sacrifice too many pawns, so to speak, but allowing it to say conduct your drone swarm for the time being, more power to to the West. So, I think it's fantastic... The techno and swarms of drones happens first before we AI the generals.”

Learning from reading is more easily automated by AI than learning from travel/experience; wise allocation of time should shift toward activities (travel, networking) that build embodied knowledge AI cannot easily replicate.

“arguably, I should travel more and read less to learn things cuz what I learn from reading, probably AI will do that quite well, right? But what I learn from traveling, it seems much harder for AI to learn that.”

Ferguson's 23-year-old son Lachlan, when unpacking Meta VR goggles that his cutting-edge tech company required employees to use for meetings, displayed 'a look of nausea and despondency' that is 'very telling' about product-market fit problems.

“my very technically minded uh, 23-year-old Lachlan was with me uh, over the past week. And the new job he has obliged him to get the goggles because the company, which is very cutting edge, wanted to start using the metaverse for meetings. Uh, the look of nausea and despondency that swept over him when he unpacked uh, the goggles was to me very telling.”

India is 'one small set of ideas away from a growth miracle' that would match China's growth trajectory, and if India liberalized policy constraints it could experience rapid development equivalent to China's recent growth.

“And they are one small set of ideas away from a growth miracle. India could have China's growth miracle tomorrow if they would only let themselves do it. Uh and that would be wonderful for the world and and geopolitics.”

Society will adapt to AI as it has adapted to Google and the internet; within 15 years, AI-integrated workflows will feel normal, though the adaptation period will be 'a big change.'

“I think we'll adapt to it. I think 15 years from now it will feel pretty fine, the way we've all adapted to a world with Google and the internet. But uh man, it's going to be a big change. I want to be ready for it.”

Ferguson's professional identity is built on 'being good at writing' for 40+ years, so AI threat to writing skills represents personal obsolescence for him; he 'takes that very personally.'

“You're talking to somebody who has to write about 2,000 to 3,000 words in the next couple of hours. Doing what I've done since I was a school boy turning my ideas into prose... And what we're talking about here is my obsolescence. And I take that very personally because being good at writing is what got me here.”

It's 'hard for me not to think' the US will benefit more from high-tech military importance than any rival, as this favors capitalist innovation-based militaries over centralized competitors.

“just if you ask on net who in military terms is most likely to benefit from high-tech being more important, it's just hard for me not to think it will be the United States.”

China is not 'expansionist power in the same way Nazi Germany was' and 'surrounded by India, Japan, possibly nuclear Japan, very hostile Australia'—regional containment coalition exists that limits Chinese sphere of influence expansion.

“I don't see them as an expansionist power in the same way Nazi Germany was. And the world will somehow have to adapt to the new geopolitical configuration, and China is surrounded by India, Japan, a possibly nuclear Japan, a very hostile Australia, still some American military presence, even if we failed in the task of defending Taiwan.”

China's stated goal in Taiwan is 'something like the Soviet Union taking over Eastern Europe after World War II'—expanded sphere of influence rather than global conquest—but still represents 'tremendous setback for US' given power concentration.

“Unlike 1939, China's not threatening to invade the United States. So, it's not quite like the situation of of the UK facing Germany. But yes, they they're they they probably will try to take Taiwan. Their goal is something like the Soviet Union taking over Eastern Europe after World War II. So, an expanded sphere of influence perhaps an axis of evil. That would be a tremendous setback for the US”

Cowen wakes up 'early in the morning' thinking 'what can I do to be ready for the changes' from AI, and does not see other public intellectuals 'asking that question every morning,' suggesting competitive advantage from constant adaptation thinking.

“And every day I wake up, and early in the morning I'm thinking, what can I do to be ready for the changes? And again, I clearly don't have good predictions, but I'm asking that question every morning. And the other people out there, I don't see that they are.”

Readers 'truly desire celebrity'—Taylor Swift big not just in music but 'incredible number of Instagram followers'—and people want 'authentic' version of celebrity/intellectual figure that AI cannot replicate, making authentic human celebrity 'luxury brand' immune to AI competition.

“I think readers truly desire celebrity. So, Taylor Swift is not just big in music, but she has an incredible number of Instagram followers. People want some version of Taylor Swift that to them at least feels authentic. I'm not saying Instagram is truly authentic, but it's authentic enough.”

Ferguson should avoid Instagram because 'your relevance as as an influencer will collapse' if you show authentic hotel bedroom—luxury brand position requires curated appearance and cannot rely on authenticity/relatable content strategy.

“Whatever you do, don't don't Instagram that that hotel bedroom, because your your relevance as as an influencer will collapse. Uh it is it is not a cool bedroom you got there.”

Both work-from-home extremism and back-to-office mandates are partially correct: remote work is viable for many senior roles and requires less coordination, but mentoring junior staff, identifying high-potential employees, and transmitting organizational culture requires in-person interaction.

“I think both of those views are true. So many workplaces, people have moved away. Uh they're not coming back, no matter what the boss may want. And it's hard to just fire them all. And indeed you don't want to just fire them all. So I think for many people, especially more senior people, work from home is fantastic. I've done work for from home for many years. I also do work from hotel room, maybe more than work from home. And that's in this sense like work from home. But that said, for mentoring young people, spotting their talent, figuring out which intern ought to someday be a senior manager, I think being in the workplace is very important.”

Three-days-per-week office policies waste real estate without delivering the coordination benefits of full-time co-location, making them economically untenable as a long-term equilibrium.

“One thing that doesn't seem plausible to me, Tyler, is the the idea that you come into the office two or three three days a week, but work from home the others, because then you're left with all the the the overhead of of office and commute, but it's getting used less. That can't be the solution.”

You could have made the same optimistic arguments about Britain in the 1920s (hyperinflation in Germany, Soviet chaos) and been proven wrong by 1939; totalitarian regimes in crisis may take strategic risks that destroy complacent Western powers, making relative strength arguments unreliable.

“But you could have said all of those things in the 1920s about Germany and the Soviet Union. You could say, 'Oh, well, you see, I know we have problems in Britain, old chap, but uh the Germans have catastrophic hyperinflation, and uh they can barely run a democracy. And as for the Soviet Union, well, the mad madness has has has put the Bolsheviks in power.' Uh fast forward to 1939, and Nazi Germany and the Soviet Union essentially take over Europe, and the only thing that really bails out the UK is that they eventually fall on one another.”

The West was eventually victorious in WWII but at catastrophic cost (British exhaustion by 1945); even if the West eventually defeats a China-Russia-Iran axis, victory could be achieved only through extreme sacrifice and loss of global primacy.

“We might end up in the end victorious as indeed the West was in 1945, but it could be a very very costly victory as it was in 1945.”

Neil Stephenson's science fiction novel Snow Crash remains 'the best visualization' of a plausible metaverse-future scenario: a stratified society with public squalor and private affluence where the underclass uses the metaverse to escape ghastly lives, and this vision is still quite plausible.

“And Neil Stephenson's Snow Crash still seems to me the best visualization of what a world with a metaverse would look like. I think Neil even coined the term metaverse. And in Snow Crash, there's clearly a a two-tier society uh in in a California of the future which has public squalor, but of course private affluence. Uh and the the underclass uses the metaverse to escape from their basically ghastly lives. And and that seems to me like a still a quite plausible vision of of where we're heading.”

Early adopters of new technologies often waste resources investing in failed platforms (C++, early video formats); picking winners is harder than predicted, making cautious skepticism about early adoption rational despite missing some winners.

“we're all old enough to see a many technologies come and go. And the early adopters spend an enormous investment on something that ends up not really being that useful. The tech I mean, just look at Zoom, which we're all looking at right now. Uh you know, Microsoft Teams is still out there. It's awful, but why it's just a little bit wrong because it's so hard to use. Uh you can you know, you invest I for a while invested a lot in C++ and turns out that was a mistake because it was a terrible language and I should have been investing in Python instead.”

Ferguson passed on opportunities to blog or engage on TikTok, instead focusing on channels where he could maintain control over quality and originality rather than being forced to produce high volume.

“I did not become a blogger. And I counseled a good friend, Andrew Sullivan, not to become a blogger because I thought it would drive him insane, and I think it very nearly did to be writing every day. You know, we talked earlier about the original originality variable. The variable I've always wanted to capture is the pages read to pages written or words read to words written ratio, where blogging encouraged people to write far too much in relation to what they were reading. Uh and I don't regret having passed on on blogging.”

GPT-3 represents primarily a reversion-to-the-mean tool and is not the obvious application for academic papers, which require originality; journalism, Twitter, and Substack are far more vulnerable to AI replacement.

“Now you have to edit which one of those you like, but it becomes a tool. I don't academic papers submitted for publication or those have to be original. So in some sense it's not the obvious application for GPT-3 which as far as I'm understanding how it works is always going to be reversion to the mean. It's going to give you kind of like the average of what would follow these things. It's better than that. Journalism, Twitter, Substack, those are just going to be 99% written by bots.”

Think tanks and institutions will face governance questions about disclosing AI contribution percentages (e.g., '83% AI') and whether to publish AI-contributed work, requiring new editorial policies.

“Hoover other institutions, think tanks will face this issue. Someone will submit an output and they'll be honest and they'll say, 'This is 83% AI.' And do you publish it? Do you Do you put their name on it? What kind of footnote is attached? Uh it will change everything.”

Immigration to California represents a form of escapism in itself—immigrants are not seeking additional escape through virtual metaverses because they are already living out the escape they made from their home countries, making them less interested in virtual reality as a refuge.

“If you think of California as a state with a disproportionately high number of immigrants who on part also make the state work, immigrants are not typically into escapism that much. They may indulge in it selectively, but they want to make their lives work. They want to give their children a better future... Already escaped, so they don't really need to do escapism.”

The metaverse may be valuable for hybrid meetings if it can equalize presence between remote and in-person participants, mimicking the moon program's unexpected spinoffs (satellite technology) by generating useful byproducts even if the primary goal (consumer metaverse) fails.

“You don't have to believe the current manifestation of the metaverse is how we'll do it, but just as the moon program gave rise to our current satellite belt, the chance that's been off from what we are calling the metaverse will be important. I think it's reasonably high.”

YouTube will prove much more important than TikTok because it serves as an educational and entertainment medium with long-term cultural persistence, whereas TikTok primarily saves time by replacing longer consumption (Netflix) with shorter bouts.

“I actually think YouTube will prove much more important than TikTok. I'm glad people enjoy TikTok. I think in some ways it saves time. People who used to waste 90 minutes watching some garbage on Netflix now spend four or five minutes on TikTok, and then they get back to their homework.”

Talent identification and support (through programs like Emergent Ventures) will be resistant to AI disruption, as the ability to identify young talent and advance it requires judgment and networks that AI cannot replicate.

“The thing I focused on in the last few years is talent identification. And we'll see how much AI disrupts that. But through my program, Emergent Ventures, we now have about 350 winners, and we support them financially, and we believe we have good means of identifying young talent and advancing it. And I think that will not be disrupted for some while.”

AI restaurant reviewing may work better than expected: an AI can learn cuisine types, read other reviews, and consult menus to produce accurate reviews without visiting; most restaurant reviewing could be done without physical presence.

“I think AI restaurant reviews will be very good soon. That's one thing it can do. Uh it is just has to read the other reviews and know something about Sichuan cuisine, be able to read the menu. I feel most of the restaurants I review, I could review them without visiting the restaurant. By looking at the menu and reading other reviews, I can be pretty accurate.”

Ferguson's skeptical response: US military advantage thesis is belied by prospect of 'aircraft carriers sink in the Taiwan straits having been overwhelmed by Chinese drone swarms' that US 'underestimated,' making confidence in US advantage overconfident.

“I'll remember those words as we watch the aircraft carriers sink in the Taiwan straits having been overwhelmed by Chinese drone swarms that we underestimated.”

Cowen would 'short Zoom' and expect improvement 'quite soon'; would also 'short TikTok in terms of influence' because reputations don't build on TikTok while YouTubers building TikTok audiences 'seeking to move to YouTube' for reputation/monetization.

“I would be short Zoom. I think it will be improved upon quite soon. And I actually would be short TikTok in terms of influence. It's striking that on TikTok, you can get an enormous number of views, but reputations don't get built in the same way as on YouTube. And people who do well on TikTok, they're often seeking to move to to YouTube cuz that's where they can internalize their reputation and monetize it.”

Green Mantle (Ferguson's organization) has successfully implemented distributed team model with regular multi-day offsites in pleasant locations, demonstrating that this approach works effectively in practice.

“That's what we've been doing at Green Mantle and it's really it's really working quite well. And it means that that we spend these quite enjoyable periods of maybe 3 days together somewhere we we choose because it's going to be a pleasant place to engage in that thing that we naked apes really need, which is to hang out together drinking alcoholic beverages to lower our inhibitions and sharing ideas informally in ways that just won't happen over Zoom.”

Quality intellectual output will remain high and abundant in an AI-mediated future, and there will be plenty of high-quality content to read and consume despite AI-generated noise, making the future 'a lot of fun' rather than dystopian.

“Quality will be high. There'll be plenty to read, plenty to listen to, and images to look at. I think it'll be a lot of fun.”

Europe has stagnated economically for a decade with no visible recovery path; comparison to Europe and UK shows the US, despite problems, retains relative dynamism, justifying cautious optimism.

“we go into this war with 100% debt to GDP. And you know, in how do you fight a war? You borrow a ton of money and you know, you good country can successfully borrow 100% maybe 120% of GDP to fight a war if they start with nothing. But if you already started 100% of GDP, your your fiscal resources to fight a sustained war are in much deeper trouble. I would agree India will never be our ally, but India is an obstacle. It's growing at over 5%. It's the world's most world's most populous nation.”

The risk of war with Iran is deeply underrated; the risk of radical transformation from AI is also deeply underrated; these are the two most under-recognized geopolitical and technological risks.

“So, what what do you think, talking about your own life for a moment, AI will transform your tremendously productive individual, uh a polymath. Uh if we fast-forward a few years, what do you think will be different about your own life? Risk of war with Iran soon, deeply underrated. Risk of positive risk of radical transformation from AI also deeply underrated.”