Tyler Cowen Says Banning Chinese Open-Source AI Will Fail
TBPN
Attempts by the U.S. government to sanction or ban open-source Chinese artificial intelligence models are bound to fail because major American companies and multinationals are already deeply integrated into their codebases.
## Background
This interview was published by TBPN on YouTube on July 20, 2026. According to the page description, it features a full interview with Tyler Cowen discussing why banning Chinese open-source AI is impossible, the rise of 'AI maniacs,' Europe's emerging AI ecosystem, job prospects in an AI economy, data center benefits, AI-generated religions, the gambling boom, and the economic 'vibecession.'
## TL;DR
In this wide-ranging discussion, economist Tyler Cowen argues that U.S. efforts to ban Chinese open-source AI models are bound to fail given deep technical integration and local deployment trends. He highlights the rise of hyper-motivated young 'AI maniacs' who are driving innovation and reshaping status dynamics across industries, particularly in regions like Europe and Estonia. Cowen also dismisses fears of AI-driven youth unemployment, addresses public skepticism toward economic indicators, explores how AI could spawn hyper-specific religious sects, and offers perspective on historical work identity and gambling policy.
## Atmosphere & style
The conversation is intellectually energetic, brisk, and conversational, striking a balance between high-level macroeconomic analysis and casual pop-culture banter. Cowen delivers sharp, counterintuitive insights with an unruffled, candid demeanor while the hosts push back with practical tech-industry perspectives and humorous anecdotes. The mood is inquisitive and engaging throughout, shifting fluidly between serious policy debate, dry wit, and collegial banter.
## Key takeaways
- [00:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=0s) Attempts by the U.S. government to ban or sanction open-source Chinese AI models are bound to fail because major firms are already heavily integrated with those codebases.
- [04:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=240s) Northern Virginia data centers fund roughly 50% of their local county's budget, demonstrating the fiscal benefits of AI infrastructure.
- [04:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=240s) Public economic pessimism or "vibecession" stems from negative emotional contagion and low institutional trust rather than weak macroeconomic indicators.
- [11:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=660s) Self-taught, highly motivated young developers ("AI maniacs") are driving model building and launching hyper-lean companies, transforming global startup ecosystems across Europe and the U.S.
- [30:35](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1835s) Concerns about AI causing immediate high youth unemployment are unsupported by data, as 18-to-24-year-old employment remains at pre-LLM levels.
- [22:51](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1371s) To maintain career advantage, workers inside traditional companies should proactively learn and integrate leading AI tools rather than relying on automated prompting.
- [19:55](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1195s) Autonomous AI agents will enable rapid creation of custom religious texts and rituals, leading to a proliferation of niche spiritual sects.
- [26:41](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1601s) Outright bans on sports betting are impractical; policy interventions should focus on tax adjustments and public health strategies similar to anti-tobacco campaigns.
- [21:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1285s) Linear demographic projections predicting widespread Amish growth are flawed because tight-knit communities face higher defect rates as total numbers scale.
- [34:53](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=2093s) Forward-thinking tech companies with strong business moats are leveraging AI-native young talent aggressively rather than solely pursuing AI-driven layoffs.
## Detailed walkthrough
### [00:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=0s) Banning Open-Source AI and Lessons from Automotive Tariffs
The conversation opens with host banter about Tyler Cowen’s coffee habits. Cowen notes that he has only consumed coffee twice in his life—both times during traditional coffee ceremonies in rural Ethiopia—reasoning that he avoids the beverage to dodge the daily dependency and headaches that afflict caffeine addicts. The discussion then transitions to artificial intelligence and the geopolitical debate surrounding open-source models, such as China's Kimi k3. Cowen argues that attempts by the U.S. government (specifically urging the Trump administration) to ban or sanction open-source AI models are bound to fail miserably, pointing out that multinational corporations and major U.S. firms already rely heavily on Chinese open-source architectures.
The hosts push back, noting that modern frontier models require specialized server infrastructure like Nvidia NVL72 racks, which could theoretically act as regulatory bottlenecks to enforce bans—far more easily than past attempts to stop digital music distribution. Cowen counters that this physical constraint is temporary, asking what regulators will do once high-powered models run locally on-device. Drawing parallels to the automotive industry, Cowen warns that if Europe engages in a trade war with China over cheap Chinese electric vehicles (like BYD), Europe will ultimately lose. He notes that protectionist tariffs only inflate input costs for domestic sectors, postponing necessary industrial adjustments rather than solving competitive weaknesses.
* Cowen has only drank coffee twice, both times during traditional rural Ethiopian coffee ceremonies.
* Multinationals and major U.S. technology companies are already deeply integrated with Chinese open-source AI codebases.
* The hosts cite Nvidia's NVL72 rack systems as potential choke points for hardware-level AI regulation.
[00:04](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=4s) Data Center Infrastructure, Tax Benefits, and the "Vibecession"
Addressing the physical AI stack—energy, server farms, chips, and networking—Cowen addresses local pushback against data center construction. Living directly adjacent to a massive cluster of data centers in Northern Virginia, Cowen praises their presence, noting that tax revenues from these facilities fund roughly half of the entire county's budget. He maintains that domestic infrastructure development and open-source models act as complements rather than substitutes: having open-source options in the ecosystem reassures enterprise customers who fear being locked into proprietary API platforms.
The hosts shift to the broader economic mood, referencing Kyla Scanlon's concept of the "vibecession"—the disconnect where macroeconomic indicators (employment numbers, stock market indexes, real wages, wealth accumulation) remain solid, yet public sentiment and consumer confidence trend overwhelmingly negative. Cowen labels this phenomenon "negative emotional contagion," a psychological loop that feeds on low institutional trust following the Great Financial Crisis and COVID-19 pandemic communications. The hosts speculate whether a brief market correction or actual recession might ironically resolve this cognitive dissonance by validating public anxiety, though Cowen cautions it could just as easily trap the public in a self-fulfilling narrative of doom.
* Data centers in Northern Virginia contribute approximately 50% of the local county's tax base.
* Historically, the tech boom from 1995 to 1998 contributed roughly 0.5 percentage points of additional growth to GDP.
* Cowen characterizes widespread public economic pessimism as "negative emotional contagion."
[00:10](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=10s) Culture, Sports Analytics, and the Odyssey Review
Turning to pop culture, Cowen shares his reaction to seeing Christopher Nolan's blockbuster film adaptation *The Odyssey*. While acknowledging it as visually dreamy and far superior to standard cinematic offerings, he considers it ultimately disappointing relative to expectations. He criticizes the dialogue as weak and calls out miscasting—specifically arguing that Matt Damon lacked the physical presence required to play Odysseus and that Calypso came across like a suburban housewife, though he defends the film's cross-ethnic casting choices. He dryly adds that classicists are simply thrilled the epic poem is back on bestseller lists and getting public attention.
When asked whether technology and data optimization are ruining sports like soccer and basketball, Cowen notes that while he does not watch soccer, quantitative analytics have undeniably homogenized professional basketball. By pushing every team toward identical tactical targets and player profiles, technology strips away distinct regional and team playing styles.
* *The Odyssey* film adaptation boosted original translations of Homer back onto consumer bestseller lists.
* Cowen highlights Matt Damon's casting as Odysseus as a prime example of the film's structural missteps.
[00:11](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=11s) The Rise of "AI Maniacs" and Europe's Emerging Ecosystem
Cowen introduces his concept of "AI Maniacs"—young, hyper-motivated teenagers and self-taught developers who spend every waking hour building models, launching hyper-lean companies, and outpacing traditional education. He recounts interviewing two such teenagers earlier that morning (a 13-year-old and a 17-year-old based in Abu Dhabi and Finland). Cowen predicts these self-taught obsessives will redefine the labor market by creating high-revenue enterprises run by extremely small headcount teams across medicine, law, banking, and consulting.
Rather than relying on government subsidies, Cowen argues that countries need cultural and institutional openness that respects young talent, allows them to buy compute, and grants them real operating authority. While the U.S. and Estonia excel on these metrics, Cowen reveals he has grown increasingly bullish on Europe over the past six months. Rather than migrating instantly to Silicon Valley (like previous generations of European tech founders such as the Collison brothers of Stripe), young European builders are utilizing open-source agentic workflows—which appeal to them more than basic conversational chatbots—to build native startups right out of Western and Eastern Europe.
* Cowen interviewed a 13-year-old from Abu Dhabi and a 17-year-old from Finland as primary examples of "AI maniacs."
* Estonia is highlighted alongside the U.S. as a prime environment for early digital infrastructure and young technical builders.
* Through his Emergent Ventures grant program, Cowen is actively funding younger, non-traditional founders to support this global shift.
[16:11](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=971s) AI Productivity Traps and Being "One-Shotted"
The discussion opens with a look at how people interact with artificial intelligence, contrasting automated "make-work" with genuine innovation. Cowen observes that younger users tend to be more playful and willing to take risks with AI models. Even if their aggregate output contains mistakes, they are ultimately the ones who will discover the game-changing applications that matter.
When the conversation turns to the phenomenon of becoming "one-shot"—getting so consumed by a single obsession that one loses touch with reality—Cowen remains unfazed, noting that people have always hyper-focused on single topics, whether it was baseball or drugs in previous generations. In comparison, being obsessed with AI is one of the better things to be "one-shot" by. One host laughingly compares AI obsession to friends who have been completely consumed by golf, spending all day studying swings and reading articles to refine their game.
- **Obsession Parallels**: Cowen compares AI hyper-focus to historical fixations like baseball, while the hosts draw a direct line to golf obsession.
- **Young vs. Older Users**: Youthful experimentation and risk-taking are viewed as the primary engine for discovering high-impact AI use cases.
---
[17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) Billionaire Reputation and Public Works
The hosts raise a question regarding billionaire public relations, referencing an ongoing $80 million campaign in California aimed at blocking a proposed billionaire tax. When asked how wealthy individuals can reshape public perception, Cowen offers direct advice: "don't apologize and keep on going." He expresses hope that the California tax proposal fails, arguing it has already harmed the state and lacks support even among many Democratic politicians.
Addressing public hostility toward the ultra-wealthy, Cowen invokes Jonathan Swift's famous maxim: *"Censure is the tax a man pays to the public for the privilege of being eminent."* He notes that high achievers across all fields—citing examples like Malcolm Gladwell, Steven Levitt, and Paul Krugman—inevitably face public backlash once they reach a certain level of success.
The hosts ask whether building tangible public monuments like parks or libraries yields better goodwill than funding diffuse, abstract initiatives like a 1% reduction in global malaria. Cowen remains skeptical that public works will win love, joking that a modern equivalent might just be buying an NFL team, or noting how the Sultan of Brunei built grand monuments without earning global affection. While he supports billionaires funding art, music, and libraries, he insists it won't make the public love them.
- **California Billionaire Tax**: An $80 million effort is active to oppose the tax; Cowen argues the tax harms California's economic standing.
- **Jonathan Swift Quote**: *"Censure is the tax a man pays to the public for the privilege of being eminent."*
- **Examples of Targeted High Achievers**: Malcolm Gladwell, Steven Levitt, and Paul Krugman.
---
[19:55](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1195s) AI-Generated Religions and Sects
Looking at early attempts to create tech-driven faiths—such as Anthony Levandowski's early efforts around 2017—the hosts inquire about the future intersection of AI and religion. Cowen predicts that AI won't necessarily create entirely new global religions, but will instead drive a massive proliferation of hyper-specific sects.
With autonomous agents, individuals will be able to generate complete religious frameworks in less than a day—complete with custom sacred texts, rituals, and lists of true imams or prophets, drawing comparisons to the historical splintering of the LDS (Mormon) Church. While most of these AI-generated sects will fail to gain traction, a few will attract hundreds or thousands of adherents. If even half of one percent of the population adopts these niche faiths, the overall religious landscape will become far more diverse, decentralized, and strange.
- **Anthony Levandowski**: Referenced for proposing an AI-focused religion as early as 2017.
- **Speed of Creation**: AI agents will enable the creation of custom holy texts, rituals, and dogmas in under 24 hours.
- **Historical Parallel**: The structural splintering of early Mormonism (LDS Church) into various sub-sects.
---
[21:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1285s) Demographics and the Amish Lifestyle Trend
The hosts bring up a demographic model projection for the year 2200, which claimed that 34% of the US population could be Amish based on simple extrapolation of birth rates. They note that the Amish way of life is gaining cultural mindshare as popular culture increasingly embraces "digital detoxing," setting screen-time limits, and telling people to "touch grass."
Having visited Amish communities many times, Cowen expresses deep admiration for their discipline and prosperity, but argues that linear demographic projections are fundamentally flawed. He points out that the Amish lifestyle is cross-subsidized by the surrounding modern society and relies on tight-knit community structures. If the population grew to three or four million, the rate of people leaving the fold would increase dramatically.
- **Projection Claim**: Simple linear extrapolation suggests a 34% Amish population in the US by 2200.
- **Structural Bounds**: Cowen emphasizes that higher total numbers would naturally lead to higher defect rates, bounding their demographic cap.
---
[22:51](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1371s) Modern Priestly Classes and Workplace AI Adaptation
Referencing a concept by Jeremy Gaffin, the hosts ask if society's search for a "priestly class" has shifted over time from billionaires to online posters. Cowen dissents, arguing that the era of the dominant online poster peaked 5 to 10 years ago, replaced by video content and direct competition from AI models themselves.
Turning to career strategy, Cowen notes that most people are not wired to be ambitious entrepreneurs—they prefer stable, lower-risk jobs in large institutions that pay better. For employees inside existing organizations who feel threatened by AI, his advice is straightforward: "Follow AI. Learn how to work with the leading models."
While tech commentators like Joe Wiesenthal argue that AI removes the learning curve by teaching users how to use it, Cowen insists that relying on the model to prompt you keeps you behind. True mastery requires staying ahead of both the tool mechanics and the broader industry developments.
- **Career Reality**: Most workers prefer structured corporate roles; Cowen urges employees to lean into "messy," AI-adjacent roles inside their companies.
- **Joe Wiesenthal View**: Noted for the opinion that AI eliminates its own learning curve by directly instructing the user.
- **Current Work Reality**: Cowen dismisses short-term expectations of an "AI leisure dividend," noting that most professionals currently have to work harder to keep pace.
---
[26:41](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1601s) The Gambling Boom and Historical Smoking Bans
The dialogue moves to the explosion of sports betting and digital gambling. Cowen views the trend with concern, calling it "already too big" and highly addictive, though acknowledging it is hard to reverse because users clearly derive entertainment from it.
The hosts highlight a paradox in gambling behavior: losers rarely blame the betting platforms or app executives; instead, they direct their anger and cyberbullying toward the individual athletes who missed a field goal or shot. When discussing potential policy interventions, Cowen opposes outright bans, favoring instead tax adjustments or playbook strategies borrowed from public health campaigns against tobacco.
This sparks a nostalgic side conversation about smoking bans. Surprised by the history, the hosts learn that smoking on short domestic flights (under two hours) was banned in 1988 before being extended to all domestic flights in 1990. Cowen contrasts smoking with gambling by pointing out that smoking created a direct negative physical externality for neighbors in enclosed spaces like airplanes, whereas someone quietly gambling on a phone on a plane causes no nuisance to the person sitting next to them.
- **Airline Smoking Ban History**: Banned on domestic flights under 2 hours in 1988, and fully banned on all US domestic flights in 1990.
- **Policy Tools**: Restricting advertising, state/local taxes, and federal adjustments to internalize negative externalities.
---
[30:35](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1835s) Labor Market Reality and the "Vibecession"
Closing out the section, Cowen analyzes current employment numbers. Despite widespread cultural anxiety about AI displacing workers, he points out that unemployment for 18-to-24-year-olds has returned to pre-LLM levels, proving that AI is not currently destroying the youth job market. While hiring has slowed down, the macro economy remains supported by strong growth in healthcare hiring.
When asked why young college graduates—particularly computer science majors from elite institutions like Stanford—feel the job market is terrible, Cowen attributes it to major sectoral shifts. The cushy, high-status tech jobs of the past are tightening their belts, while new opportunities lie in areas that educated elites may not find appealing. As "AI maniacs" experience a massive upward leap in economic power, society will face a major reallocation of status that traditional politics will struggle to navigate.
- **Youth Unemployment**: Rates for 18-to-24-year-olds sit at pre-LLM levels.
- **Hiring Anchors**: Healthcare hiring serves as a primary stabilizer in the broader employment data.
- **Status Shift**: A mismatch exists between educated graduates seeking traditional corporate roles and the fast-growing opportunities being captured by dedicated AI builders.
[32:53](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1973s) Skill Issues, Hiring Strategy, and Work Identity
The conversation opens with a discussion around recent graduate job placements, specifically addressing concerns regarding Stanford computer science grads. A team member named Tyler suggests that struggling to find work with such a credential is a "skill issue." The sentiment is that top-tier graduates can readily secure a job or launch a startup, though they may need to adapt their mindset and consider non-traditional career tracks. Tyler Cowen agrees with this assessment, acknowledging that the path to employment in tech is shifting.
The hosts shift to discussing the optics and strategy surrounding corporate workforce decisions. While many CEOs are implementing AI-driven layoffs, the hosts argue that companies with durable business models and strong moats should instead go on the offensive. They point to Cloudflare CEO Matthew Prince hiring around a thousand interns as a prime example of leveraging young talent who understand AI to amplify business productivity. While companies coming out of a previous hiring glut may need to retool, tech and AI firms themselves continue to hire aggressively.
Finally, the dialogue touches on the psychological and historical relationship between work and human identity. When asked if there was ever a period in history where the majority of people did not derive their self-worth and identity from their jobs, Cowen immediately points to medieval times—noting that while people spent their lives growing food, their sense of self operated in a radically different framework than today's employment-centric culture. Brief audio stutters punctuate the exchange, prompting jokes that Cowen's internet connection briefly traveled back to 1988.
- **Graduate Market:** Stanford computer science grads face shifting job tracks, requiring a mindset shift rather than facing a complete lack of opportunity.
- **Strategic Hiring:** Companies with durable business models should hire AI-native talent (referencing Matthew Prince hiring ~1,000 interns at Cloudflare) rather than pursuing defensive AI layoffs.
- **Historical Work Identity:** In medieval times, work like farming was central to survival, but individual identity and self-worth were not structured around modern concepts of career and employment.
[35:00](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=2100s) Wrapping Up and Inviting Tyler to the Studio
As the interview comes to a close, the hosts express their gratitude to Tyler Cowen for a wide-ranging conversation covering topics from around the world. Wishing him a great week and summer, they extend an open invitation for him to visit their studio in California the next time he is in the area. They joke about attempting to host a four-hour marathon episode over "the most memorable cup of coffee in history" and express confidence that peer pressure will successfully convince him to return. Cowen warmly bids farewell as the interview wraps up.
## Notable quotes
> [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) don't apologize and keep on going — Tyler Cowen
> [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) Censure is the tax a man pays to the public for the privilege of being eminent. — Jonathan Swift
> [22:51](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1371s) Follow AI. Learn how to work with the leading models. — Tyler Cowen
## Mentioned
Tyler Cowen, TBPN, China, Kimi k3, U.S., Donald Trump, Nvidia, NVL72, Europe, BYD, Northern Virginia, Kyla Scanlon, Christopher Nolan, The Odyssey, Matt Damon, Calypso, Homer, Abu Dhabi, Finland, Estonia, Patrick Collison, John Collison, Stripe, Emergent Ventures, California, Malcolm Gladwell, Steven Levitt, Paul Krugman, Sultan of Brunei, NFL, Anthony Levandowski, LDS Church, Amish, Jeremy Gaffin, Joe Wiesenthal, Stanford University, Matthew Prince, Cloudflare, Ramp, Public, Cisco, Console, CrowdStrike, Figma, MongoDB, NYSE, Railway, Shopify, Codex, OpenAI
泰勒·考恩称封禁中国开源AI注定失败
美国政府制裁或封禁中国开源人工智能模型的企图注定会失败，因为美国大型企业和跨国公司已与其代码库深度融合。
在这场广泛的讨论中，经济学家泰勒·考恩（Tyler Cowen）指出，鉴于深度的技术整合与本地部署趋势，美国封禁中国开源AI模型的举措注定失败。他强调了一批极具动力的年轻“AI狂热者”的崛起，他们正推动创新并重塑各行业的地位格局，在欧洲和爱沙尼亚等地区尤为显著。此外，考恩还驳斥了对AI引发青年失业的担忧，回应了公众对经济指标的质疑，探讨了AI可能如何催生极度细分的宗教派别，并对历史职业身份认同与博彩政策提出了见解。
Tyler Cowen 认为美国政府封禁或制裁中国开源 AI 模型的企图注定失败，因为大型跨国企业与美国主流科技公司早已深度集成了这些代码库。 (0:00)
弗吉尼亚州北部的数据中心集群为当地贡献了约 50% 的财政税收，印证了 AI 硬件与能源基础设施为地方带来的显著财政红利。 (4:00)
被称为“AI 狂人”（AI Maniacs）的年轻自学开发者正借助开源智能体工作流，在欧洲和美国打造极度精简但高营收的初创企业，重构全球创业生态。 (11:00)
自主 AI 智能体使在 24 小时内生成专属经文和仪式成为可能，未来或将催生大量高度细分的小众精神教派与分支。 (19:55)
关于 AI 引发青年严重失业的恐慌缺乏数据支撑，当前 18 至 24 岁群体的就业率仍保持在 LLM 兴起前的水平，宏观就业依然受到医疗等行业招聘的稳固支撑。 (30:35)
Reading summary
Tyler Cowen argues that attempts by the U.S. government to ban foreign open-source AI models are unworkable and economically harmful, as multinational enterprises already rely on them. He suggests that rather than precipitating broad labor collapse, AI is triggering a cultural and economic status reallocation favoring self-taught young builders over traditional corporate credentials.
Open-Source AI and Infrastructure Realities (0:00)
Cowen asserts that efforts by U.S. policymakers to restrict or sanction Chinese open-source architectures, such as Kimi k3, will fail because multinational corporations and major domestic technology firms have already integrated these models into their operations. While hosts suggest that physical choke points like Nvidia NVL72 server racks could enable hardware-level enforcement, Cowen contends this constraint is temporary, anticipating that high-capacity models will soon run locally on-device. Comparing the situation to European trade disputes over low-cost Chinese electric vehicles like BYD, Cowen warns that protectionist tariffs only elevate domestic input costs and delay necessary structural adaptations.
Domestically, Cowen defends the aggressive buildout of physical AI infrastructure. Citing data center clusters in Northern Virginia that generate roughly 50% of his local county's tax base, he views physical data facilities and open-source models as complementary rather than mutually exclusive, noting that open alternatives reassure enterprise clients wary of proprietary API vendor lock-in.
The 'Vibecession' and Macroeconomic Sentiment
Addressing the 'vibecession'—the gap between strong macroeconomic indicators like employment and real wages versus persistent public pessimism—Cowen diagnoses the issue as 'negative emotional contagion.' He attributes this psychological loop to declining institutional trust stemming from the Great Financial Crisis and official communications during the COVID-19 pandemic. While some speculate that a market correction might resolve this dissonance by validating public worry, Cowen cautions that an actual downturn could instead cement a self-fulfilling narrative of decline.
Labor Market Realities and 'AI Maniacs'
Despite widespread anxiety over technological displacement, Cowen notes that unemployment for 18-to-24-year-olds sits at pre-LLM levels, stabilized by strong hiring in sectors such as healthcare. Elite graduates, including Stanford computer science majors, face friction primarily because corporate tech firms are scaling back previous over-hiring. Rather than a total absence of opportunity, the discussion frames this as a mindset and status shift, with companies like Cloudflare going on the offensive by hiring roughly 1,000 interns to leverage AI-fluent young talent.
Cowen highlights the emergence of 'AI maniacs'—obsessive, self-taught teenagers who build lean, high-revenue companies that bypass traditional credentials across medicine, consulting, and law. Pointing to builders in Abu Dhabi, Finland, and Estonia, Cowen observes that young European founders are increasingly using open-source agentic workflows to launch native ventures at home rather than emigrating to Silicon Valley. For legacy corporate workers, Cowen dismisses expectations of a near-term leisure dividend, advising professionals to directly master leading models rather than waiting for automated prompting to guide them.
Cultural Fragmentation, Gambling, and Wealth Backlash (17:28)
Cowen forecasts that generative AI will accelerate cultural fragmentation through the rapid creation of niche religions. With autonomous agents capable of drafting custom sacred texts and dogmas in under 24 hours, he predicts a splintering pattern similar to early Mormonism, producing hundreds of minor sects that will diversify society even if adopted by only a fraction of a percent of the public. Conversely, he rejects linear demographic forecasts claiming the Amish will comprise 34% of the U.S. population by 2200, noting that community expansion faces structural defect caps and relies on cross-subsidies from the surrounding modern economy.
Addressing other policy challenges, Cowen expresses concern over the expansion of digital sports betting, viewing it as highly addictive. Rather than implementing outright bans, he favors tax adjustments and advertising restrictions analogous to historical anti-tobacco campaigns, distinguishing phone gambling from the 1988 and 1990 airline smoking bans because it lacks immediate physical externalities. Regarding hostility toward the ultra-wealthy—highlighted by an $80 million campaign opposing a California billionaire tax—Cowen quotes Jonathan Swift's maxim that censure is the tax paid for eminence, maintaining that public works and civic philanthropy will not shield high achievers from societal resentment.
While Cowen outlines an adaptive economic future driven by physical infrastructure and ambitious young developers, his analysis leaves open how traditional governance and political systems will absorb the resulting status displacement, negative public sentiment, and cultural fragmentation.
## Background
本期访谈由科技播客频道 TBPN 于 2026 年 8 月 3 日发布在 YouTube 平台。节目简介显示，本期为对知名经济学家泰勒·考恩（Tyler Cowen）的完整专访，重点探讨了美国为何无法封禁中国开源 AI 模型、“AI 狂人”的涌现、欧洲悄然崛起为 AI 重镇、数据中心与未来就业市场的演变，以及 AI 催生的新兴宗教派系和经济“情绪衰退”（vibecession）等多元议题。
## TL;DR
泰勒·考恩在对话中鲜明指出，通过行政手段封禁中国开源大模型是不切实际的幻想，开源模型不仅因端侧化演进而难以监管，更与美国专有闭源大模型形成了互补共生关系。他观察到全球正涌现出一批极年轻的“AI 狂人”，推动着员工极少却营收极高的轻量化企业颠覆传统行业，且欧洲青年在 AI Agent 浪潮中的活力打破了既往悲观预期。在就业与经济层面，考恩反驳了 AI 摧毁初级就业的恐慌，指出青年失业率已回落至大模型普及前的水平，当下的阵痛实为传统受教育精英阶层遭遇的行业结构性转移与社会地位再分配。此外，他还深入剖析了“情绪衰退”背后的负向心理传染、博彩成瘾难以靠一纸禁令解决的现实阻碍，以及 AI 工具极速降低门槛对未来宗教派系分化的深远影响。
## Atmosphere & style
整场对话在硬核深度的经济学洞见与幽默风趣的即兴闲聊之间自如穿梭，节奏紧凑且充满智识火花。从开场考恩透露自己一生仅在埃塞俄比亚喝过两次咖啡的极端自律，到锐评《奥德赛》电影选角以及将沉迷 AI 与高尔夫球瘾相提并论，两位主持人的敏锐追问与考恩标志性的逆向思维碰撞出极具张力的讨论。聆听过程中既能感受到考恩面对技术恐惧与宏观悲观论调时泰然自若的理性底色，又能体会到充满朋友间调侃、技术吐槽与真诚探讨的松弛谈话质感。
## Bulletin
- [01:34](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=94s) 知名经济学家泰勒·考恩（Tyler Cowen）呼吁美国政府放弃封禁中国开源大模型的企图，指出开源模型在端侧运行将使硬件层面的监管杠杆失效，且开源与闭源实际上是互补共生关系。
- [03:13](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=193s) 考恩断言若欧洲与中国爆发全面贸易战欧洲必输无疑，因为欧洲缺乏可转换的优势产业纵深，筑起关税壁垒只会推迟结构性改革并削弱制造业整体竞争力。
- [30:35](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1835s) 18 至 24 岁年轻群体的失业率已回落至大语言模型爆发前的水平，反驳了 AI 正在摧毁初级就业市场的论调；当前劳动力市场主要处于受教育精英与“AI 狂人”之间的行业结构性转移和地位重组。
- [19:44](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1184s) 借助 AI Agent 极速生成典籍与教规的能力，未来宗教生态将更趋去中心化和碎片化，从传统主流宗教分化出海量的细分派系而非全新的大一统宗教。
## Key takeaways
- [01:34](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=94s) 封禁中国开源大模型是不切实际的行政干预，跨国企业合规困难，且随着模型向端侧（On-device）演进，依托数据中心硬件算力的物理监管杠杆将彻底失效。
- [04:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=265s) 开源大模型与美国专有闭源模型在商业上属于互补品，开源方案为全球用户提供了防范断供制裁的安全后盾，从而降低了客户接入美国闭源生态的心理门槛。
- [03:13](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=193s) 面对中国电动车冲击，欧洲缺乏产业纵深无法轻松进行赛道切换，若打全面贸易战必输无疑，关税壁垒只会逐步推迟必要的结构性改革。
- [06:16](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=376s) 美国联邦制为数据中心等大型基础设施建设提供了制度弹性与冗余度，即使局部民意阻力较大，科技企业依然可以通过跨州协商落地算力与电力项目。
- [07:06](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=426s) “情绪衰退”（Vibecession）本质上是一种负向情绪传染，民众对宏观机构普遍存在幻灭感，导致主观悲观情绪与稳健的硬性经济指标出现脱节。
- [09:37](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=577s) AI 对宏观经济生产力的拉动过程将类似 1990 年代中后期的互联网浪潮，虽然不会瞬间翻天覆地，但随着时间推移终将带来显著的经济增速提速。
- [11:45](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=705s) 极少数自学成才的“AI 狂人”（AI maniacs）将借助顶尖模型与端到端 Agent 创办超轻量、高营收的企业，对医疗、法律、金融等传统专业服务业发起冲击。
- [14:53](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=893s) 欧洲年轻一代在构建实用型 AI Agent 方面展现出逆袭势头，并且凭借爱玩天性避开了成年人容易陷入的“假工作陷阱”和工具型空壳流程。
- [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) 公众对超级富豪的非议属于显赫地位的“批判税”，强推亿万富翁税会重创区域经济，单纯修建实体纪念碑或进行抽象慈善都无法直接逆转负面审视。
- [19:44](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1184s) AI 大幅降低了创教与编写教义的门槛，未来宗教景观将演变为去中心化、高度细分的小众派系林立，打破传统宗教权威的垄断。
- [23:21](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1401s) 大众创业对绝大多数普通人而言并非现实路径，身处大机构的职场人必须主动高频实践最顶尖的模型，不能被动等待所谓“休闲红利”的降临。
- [26:41](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1601s) 强行全面禁止体育博彩在实操中不可行，与控烟运动相比，手机博彩缺乏直观可感知的感官危害和二手伤害，负外部性更加隐蔽。
- [31:47](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1907s) 青年失业率指标证伪了 AI 摧毁初级就业的恐慌，当前就业痛感主要来自受教育精英阶层遭遇的地位重组与行业结构性转移。
- [33:15](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1995s) 不少企业借 AI 转型之名裁员实为掩盖此前盲目扩张的过度招聘，真正具备业务护城河与野心的企业反而会逆势扩张以利用 AI 工具放大杠杆。
## Detailed walkthrough
### [00:06](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=6s) 咖啡轶事与极端克制的晨间习惯
节目一开始，主持人迎来了首位嘉宾——美国知名经济学家泰勒·考恩（Tyler Cowen）。在常规问候之后，考恩立刻抛出一句出人意料的开场白：“我不喝咖啡。”这让正打算循例提问的主持人措手不及。考恩透露，自己这一生只喝过两次咖啡，而且两次都是在埃塞俄比亚乡村的传统咖啡仪式上。当时的体验极其美妙，以至于他当场认定：“这就是巅峰了，我不可能再尝到比这更好的了，因此无法被超越。”
面对主持人半开玩笑地追问这究竟是怕自己变得“过于强大”，还是一场“浮士德式的交易”，考恩给出了非常务实的解释：他目睹了太多人对咖啡因极度成瘾，一旦早晨喝不上咖啡就会头痛欲裂，他绝不想陷入这种生理依赖。虽然主持人调侃咖啡早已演变成现代人必不可少的供应链，且从经济学角度看，每颗仅需1美分的散装咖啡因药片具有极高的经济效率却无人问津，人们更偏爱花费数美元去买一杯咖啡带来的简单快乐。但考恩坚持认为，在旅途中或酒店里为了喝一杯劣质咖啡而大费周章是极其折磨人的，远离这种成瘾性习惯生活反而简单得多。当被问及早晨是否有任何雷打不动的仪式感时，考恩打趣说自己喝矿泉水就能“上头”（get high），但就算没有也完全不会感到渴望。
- 考恩平生仅喝过两次咖啡，均在埃塞俄比亚乡村咖啡仪式中饮用。
- 主持人提到散装咖啡因药片与单杯咖啡效果相当，成本仅约1美分，但在消费市场几乎完全没有替代普通咖啡的普及率。
---
### [01:34](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=94s) 封禁中国开源大模型是不切实际的幻想
话题迅速切入人工智能领域，主持人询问考恩对以月之暗面 Kimi k3 等为代表的中国开源 AI 模型引发的争议有何看法。考恩旗帜鲜明地指出，无论人们喜欢与否，开源大模型不仅已经到来，而且正席卷整个行业，其中大量成果必然来自中国。美国应当允许本土竞争对手在公平的起跑线上自主推进，而绝不应该诉诸行政手段。考恩直言，任何试图通过立法取缔、直接封禁或施加制裁来遏制开源模型的企图都将“惨败收场”（fail miserably）。他质问那些主张封禁的人：跨国企业该如何合规？那些海外分支机构使用了开源模型的企业怎么办？更现实的是，目前大量美国顶尖科技巨头本身就在深层代码中使用中国开源技术。考恩甚至直接向特朗普政府公开呼吁：“我希望并恳请特朗普政府放弃这场十字军东征，因为这根本不可能行得通。”
主持人对此提出了反驳，认为虽然 AI 权重在软件层面就像音乐文件一样可以无限复制并全球扩散，但目前顶尖大模型的运行严重依赖极其昂贵的实体基础设施——比如英伟达（NVIDIA）的 NVL72 机架式服务器集群。这类物理设备可以成为监管的绝佳杠杆点，因此封禁大型开源模型在实操上甚至可能比当年的反数字音乐盗版更容易。考恩随即用一个极具前瞻性的反问击碎了这一论点：“当这些模型能在端侧设备（On-device）上直接运行时，你又该怎么办？距离那一天还能有多久？既然如此，为什么一开始要走上封禁这条死路？”
```mermaid
flowchart TD
accTitle: 开源AI生态与封禁逻辑冲突
accDescr: 展示试图封禁中国开源AI模型所面临的硬件杠杆失效与生态共生关系
subgraph Hardware["硬件与基础设施层"]
A["高算力集群需求<br/>(NVL72 机架服务器)"] -->|算力监管抓手| B["短期管控假设"]
C["端侧模型演进<br/>(On-Device AI)"] -->|摆脱中心化算力| D["物理监管失效"]
end
subgraph Ecosystem["全球软件与产业生态"]
E["中国开源大模型"] -->|底层技术集成| F["美国大型企业与产品"]
E -->|降低锁定风险的备用项| G["美国专有闭源模型"]
G -.->|互补共生| E
end
B -.->|技术迭代| D
D --> H["行政封禁注定失败"]
F --> H
```
---
### [03:13](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=193s) 汽车关税的前车之鉴与欧洲工业的困境
主持人顺着全球竞争的话题，引出汽车行业在中美欧博弈中的历史教训。欧洲目前试图通过关税来应对大量涌入的比亚迪（BYD）等中国电动车，但欧洲传统车企不仅面临本土失守的风险，更在失去中国这个庞大市场的市场份额，本土替代品牌的崛起让其销量持续走低。
考恩对此做出了悲观预测，断言“如果欧洲与中国爆发全面贸易战，欧洲必输无疑”。他认为这反映了欧洲当前令人叹息的经济脆弱性：欧洲根本没有能力说出“少造几辆车，转而投身于其他五六个我们占据优势的伟大产业”这样的话，因为欧洲根本没有可供随时转换的产业纵深。简单地筑起高额关税壁垒不仅无法帮助欧洲脱困，反而会推迟必要的结构性改革。更致命的是，关税在长期会逐步传导并渗透到欧洲其他制造业部门的上游原材料与中间投入品中，进一步削弱欧洲整体经济的全球竞争力。考恩强调，欧洲现在陷入了进退维谷的绝境，虽然没有轻松的出路，但他依然坚决反对采取关税壁垒的做法。
---
### [04:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=265s) AI全栈生态位：为什么开源模型与闭源巨头是互补品
在讨论整个 AI 产业链（涵盖能源、数据中心、芯片制造及高速网络）时，主持人提出担忧：美国在本土面临重重阻力的情况下依然疯狂兴建数据中心，但如果未来开源大模型主要由中国提供，是否会导致一种滑坡效应——人们会觉得既然开源算法在中国，那么高能耗、重污染的数据中心和发电厂干脆也放到海外，从而让美国逐步丧失对整个技术栈的掌控力？
考恩以自己居住的弗吉尼亚州北部为例进行了驳斥。该地区是全球数据中心最为密集的区域之一，考恩不仅毫无怨言，反而高度赞扬这种集聚效应，指出当地数据中心贡献了所在郡县一半的税收。他对于算力、芯片和电力的建设持完全开放态度：“全部放马过来吧（bring it on）。”考恩深刻指出了开源与闭源之间的共生经济学：全球用户和企业普遍对美国专有闭源大模型抱有一种潜在恐惧，生怕未来的某一天这些 API 会被突击断供或被美国政府以制裁为由强行夺走。正因为有中国开源模型作为随时可替代的安全后盾，全球客户才更愿意在一定程度上锁定并接入美国的闭源系统。考恩强调：“这两者在事实上是互补品（complements）。”许多美国大型科技公司目前就在底层深度使用中国开源模型，主持人也证实，一些初创企业确实直接利用中国开源模型来冷启动自家产品的关键研发步骤，两者在供应链内部已经无法分割，根本无法像做外科手术一样将其“生生剥离出来”。
- 弗吉尼亚州北部的数据中心为当地县政府贡献了高达 50% 的税收。
- 主持人提到企业级用户为缩减成本，直接在生产管线中采用中国开源模型；像 Thinking Machines 等企业也曾借力中国开源模型来完成关键产品步骤的冷启动。
---
### [06:16](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=376s) 数据中心的民意阻力与美国的制度韧性
针对数据中心虽然经济收益显著但在民意调查中支持率惨淡的现象，主持人探讨这种阻力究竟是源于对人工智能取代工作的恐慌，还是地方现实生活受打扰带来的反弹。
考恩分析认为，当下的阻力源自公众对几乎所有事物普遍存在的负面心理。当下民众对国家前景、除军队以外的所有社会公共机构都抱有极高的不信任感。在经历了 2008 年金融危机和 COVID-19 疫情大流行之后，民众多次在危机时刻被官方告知“一切安好”，但事后证明事实并非如此，因此民众的幻灭和疑虑完全情有可原。但考恩对美国的制度机制保持乐观：美国是一个由 50 个州组成的联邦制国家，制度拥有强大的冗余度和灵活性。“如果马克·扎克伯格（Mark Zuckerberg）需要建数据中心，他完全可以直接给路易斯安那州州长打电话，谢天谢地这种沟通机制在我国切实有效。”只要有竞争性的州愿意提供土地与电力，数据中心的建设就不会真正停滞。
---
### [07:06](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=426s) 情绪衰退（Vibecession）与负向情绪传染
主持人将话题引向了由金融博主凯拉·斯坎伦（Kyla Scanlon）创造的热词——“情绪衰退”（Vibecession）。现实中呈现出一种违背直觉的悖论：宏观经济在稳步增长，就业市场保持稳定，当民调询问邻居的生活状况时大家都认为别人过得还行，但一问到自身情况和整体经济时，消费者信心指数却跌入谷底，人人都觉得自己过得很惨。
考恩用他自己的概念将其定义为“负向情绪传染”（negative emotional contagion）。这种现象在世界历史上多次出现，一旦形成就会自我强化并不断滋生负面预期。虽然起因可能只是某些具体的负面事件，但一旦开始蔓延就极难逆转，历史上往往需要一场真正的战争或毁灭性灾难才能打破这种心理循环。考恩指出，从硬性经济指标来看——包括财富积累、就业市场活跃度、股票价格以及实际工资水平——美国的表现都在“尚可到良好”的区间，绝对谈不上糟糕。
对于这种情绪扭曲，主持人提出了一种略带反讽色彩的假说：既然民众的心理预期已经差到了极点，那么一次短暂的经济衰退或市场剧烈回调反而可能让美国展现出更强的韧性。因为人们会觉得“我早就料到会这样，最坏的结果终于靴子落地了”，这种市场现实与主观情绪之间的认知失调一旦消除，人们反而敢于重新投资并重整旗鼓。考恩认为这种假说有其合理性，甚至可能成为一种看涨信号，但他同时也发出了警示：如果真正可怕的危机爆发，全社会很可能会陷入“我早就警告过你们”的抱怨中，从而彻底深陷泥潭无法自拔。话题最后，主持人谈及考恩一年前发表的博文——考恩当时断言 AI 真实且不可思议，但经济仍将维持原有的温和增长轨迹——并借此追问，过去一年的技术进展是否促使他修正了自己的世界观。
### [09:37](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=577s) AI 带来的生产力增长与电影《奥德赛》锐评
谈及对人工智能技术影响的预期，泰勒·考恩（Tyler Cowen）指出，许多人在社交平台或群聊中曾设想：如果如今拥有了像 GPT-5 这样强大的模型，世界本该发生天翻地覆的改变，但现实并非如此。考恩回顾了 1995 年至 1998 年的互联网浪潮，当时的技术进步为经济额外带来了大约 0.5 个百分点的增长。在他看来，那时的技术变革甚至不如当前的 AI 令人惊叹；因此，尽管进展可能会相对缓慢，但随着时间推移，AI 终将带来同等甚至更高的增长提速。
随后话题转向了备受关注的电影《奥德赛》（The Odyssey）。考恩直言，虽然该片梦幻而震撼，但整体而言令人失望。他批评对白过于乏味，选角也存在严重偏差：尽管他不介意跨族裔选角，但演员气质明显不合拍——马特·达蒙（Matt Damon）在体魄上完全缺乏奥德修斯应有的威严；扮演卡吕普索（Calypso）的女演员看起来像个郊区家庭主妇；而赞达亚（Zendaya）虽然表现不错，却没有获得实质性的角色分量。尽管如此，考恩依然认为这部三小时的长片值得一看。当被问及古典学者是否会对影片感到满意时，考恩打趣道，古典学界只要能分到一点大众关注度就喜出望外了，随着各版本《奥德赛》重回畅销书榜单，“他们实在没什么好抱怨的”。
在谈到科技对体育竞技的影响时，考恩承认自己虽然不看足球，但科技在某种程度上确实毁了篮球：算法和数据分析让比赛趋于同质化，所有球队都在追逐相同的指标和相同类型的球员，从而剥夺了各支球队独具一格的风格特色。
---
### [11:45](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=705s) 崛起的“AI 狂人”与初创生态
考恩引出了一个核心概念——“AI 狂人”（AI maniacs）。他分享了当天早晨刚交谈过的两位少年：一位来自阿布扎比，年仅 13 岁；另一位来自芬兰，只有 17 岁。这些年轻人将所有课余时间都投入到钻研 AI 之中，其掌握的知识已经远远甩开了学校老师。他们正在自主训练模型并创办公司，敏锐地意识到这才是未来的走向。考恩预测，未来商业界将涌现大量员工极少、但营收极高的超轻量型企业；只要规则允许，这群“AI 狂人”将在医疗、法律、商业、咨询和银行等各行各业发起冲击。他甚至直言，英国能否实现经济复苏，本质上取决于英国是否愿意鼓励本国的“AI 狂人”。
对于年轻人是否需要额外激励，主持人和考恩展开了讨论。考恩强调，技术的易用性并不等同于成功水到渠成，社会必须给予年轻人应有的尊重和权威地位，而不是轻视他们，同时要让他们能够买得起算力并获得资金资助。美国和爱沙尼亚在这一指标上表现亮眼，但世界上很多地区由于规模太小或文化过于保守而无法适应这种节奏，这就像虽然摇滚乐技术门槛不高，但并非所有国家都能孕育出伟大的摇滚乐一样。
```mermaid
flowchart LR
accTitle: AI Maniacs Growth Path
accDescr: The journey of young AI maniacs from self-learning to building high-revenue, lean enterprises.
subgraph S1["自我驱动与探索"]
A["年轻自学者<br/>(13-17岁少年)"] --> B["掌握前沿工具与模型<br/>(知识超越传统教师)"]
end
subgraph S2["外部要素赋能"]
B --> C["算力支持与小型资助<br/>(跨越VC死亡之谷)"]
end
subgraph S3["商业落地形态"]
C --> D["轻量化高营收企业<br/>(极少员工+高额收入)"]
D --> E["颠覆传统专业服务<br/>(医疗、法律、金融等)"]
end
```
- **极简团队模式**：AI 使得极少数年轻人依靠高级模型即可运作高流水企业，重塑传统服务业竞争格局。
- **自学成才为主**：除了少数顶尖 AI 实验室和科技巨头，传统大学几乎无法传授这些前沿技能，自我探索是唯一路径。
- **资助逻辑变化**：考恩在推进资助计划时，明确表示自己的资助对象正朝着“更年轻、更奇特”（younger and weirder）的方向演进。
---
### [14:53](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=893s) 欧洲的逆袭与年轻人对“假工作陷阱”的免疫
欧洲在“AI 狂人”这一维度上的亮眼表现让考恩感到惊喜。长期以来，欧洲顶尖科技人才外流至美国创业（如爱尔兰的 Collison 兄弟），本土除少数企业（如瑞典的 Spotify）外鲜有代表。但考恩指出，欧洲对这波浪潮的反应与以往不同：聊天机器人（Chatbots）对欧洲年轻人吸引力有限，因为纯语言对话似乎是在与悠久文化竞争；而 AI Agent（智能体）则提供了一个切实去行动、构建工具的全新舞台。从西欧到东欧，涌现出了大量年轻创业者，这让考恩对欧洲未来前景的乐观程度远超半年前。
针对部分成年人在使用 AI 时容易陷入“假工作陷阱”（例如构建复杂的自动化会议总结流程、打造从不翻看的第二大脑，产出大量工具型空壳）的现象，考恩认为年轻人能够凭借天然的爱玩天性避开这些误区。年轻人在探索中或许在整体成功率上会遭遇更多失败，但在孕育真正具有突破性的事物方面，他们注定会胜出。当被问及深度沉迷 AI 是否会导致“走火入魔”（one-shotting）并脱离现实时，考恩幽默地反驳道，历史上人们沉迷过棒球、甚至滥用药物，如今就算年轻人把精力全部“单押”在 AI 上，也比沉迷于其他不良嗜好要好得多；主持人也笑称，这跟那些整天钻研挥杆、彻底被高尔夫球绑架的中年人本质上毫无区别。
---
### [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) 富豪税争议与公信力的“批判税”
围绕加利福尼亚州正在讨论的“亿万富翁税”，主持人提到反对阵营已投入约 8000 万美元开展公关行动，并向考恩请教超级富豪应如何扭转公众的负面看法。考恩给出了直截了当的建议：“不要道歉，继续前行。”他认为加州的富豪税一旦通过，将会彻底毁掉加州经济，事实上就连加州许多民主党政客也对其持反对态度。公众对富豪的厌恶很大程度上源于弥漫全社会的负面情绪，即便某些富豪确实存在招人厌烦之处，也是人之常情。
考恩引用了乔纳森·斯威夫特（Jonathan Swift）的名言来概括这种社会现象："Censure is the tax a man pays to the public for the privilege of being eminent"（指责是显赫之人为了享有卓越地位而向公众缴纳的税款）。无论是在商业还是知识界，从马尔科姆·格拉德威尔（Malcolm Gladwell）、史蒂芬·列维特（Steven Levitt）到保罗·克鲁格曼（Paul Krugman），凡是取得卓越成就的人都会招致无端非议。
对于富豪是否应通过修建实体公共工程（如城市公园、图书馆）来重塑公众形象，考恩与主持人探讨了抽象慈善与实体丰碑的心理差异：
- **抽象影响 vs. 可视地标**：资助消灭疟疾降低 1% 发病率固然功德无量，但对大众而言过于抽象渺茫；而一座伫立在街角的大型图书馆或公园，却能带来直观的感官震撼。
- **现代意义的丰碑**：主持人笑称，如今富豪打造现代“丰碑”的方式或许已经演变成买下一支 NFL 球队。
- **丰碑的局限性**：考恩对此反问道，文莱苏丹也为民众修筑了宏伟的丰碑，但这并未让他因此赢得更广泛发自内心的爱戴，单纯建造实物并不能简单消除外界的审视与非议。
### [19:44](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1184s) 亿万富翁的声誉与 AI 催生的新兴宗教派系
访谈开篇延续了对富豪阶层的讨论。Tyler Cowen 指出，虽然他十分赞成当今的亿万富翁多出资支持艺术、音乐和图书馆等公益事业，但这种慈善行径绝不会真正让他们受人爱戴。主持人顺势将话题引向宗教与 AI 的结合，提到早在 2017 年 Anthony Levandowski 就创立了崇拜人工智能的宗教雏形，并提问为何在如今模型能力如此强大的背景下，尚未见到成熟且成体系的 AI 宗教涌现。
Cowen 认为未来的趋势并非产生前所未有的全新宗教，而是分化出海量的细分派系（sects）。借助 AI Agent，任何人都能在一天之内编写出专属的印度教典籍，或是设定一套属于自己的什叶派伊斯兰教及合法的伊玛目传承谱系，并配备完整的仪式与教规。这就像历史上摩门教（后期圣徒运动，LDS Church）派系林立的演变过程一样。尽管大部分自创宗教无人问津，但只要有千分之五的人口尝试，且部分派系聚集数百或数千名信徒，整个社会的宗教生态就会彻底演变，变得更加多元、离奇，且不再被中心化的传统宗教权威所垄断。
```mermaid
flowchart TD
accTitle: AI推动宗教派系分化
accDescr: 展示了AI工具如何快速降低创教门槛并导致宗教景观去中心化与碎片化的过程
subgraph Input ["技术赋能"]
A["AI Agent 工具"] --> B["极速生成典籍与仪式<br/>（单日内即可完成）"]
end
subgraph Process ["传统宗教派生"]
B --> C1["自订版本印度教典籍"]
B --> C2["自创什叶派传承谱系"]
B --> C3["类似早期 LDS 摩门教的分裂派系"]
end
subgraph Outcome ["长远社会影响"]
C1 & C2 & C3 --> D["聚集数百至数千小众信众"]
D --> E["去中心化、碎片化且高度多元的宗教景观"]
end
```
### [21:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1285s) 阿米什人生活方式的流行悖论与“线上祭司阶层”
主持人提到一项利用 AI 外推至 2200 年人口的预测，称未来美国将有 34% 的人口是阿米什人（Amish），并讨论起当前大众文化对阿米什生活方式的推崇——远离屏幕、限制社交媒体、提倡“去户外接触大自然（touch grass）”。对此，Cowen 明确反驳了这一人口预测。他多次拜访阿米什社区，非常钦佩其成就，但指出阿米什模式之所以能存续，是因为外部现代社会对其进行了交叉补贴（cross-subsidization）。如果阿米什人口真扩张到几百万的规模，其紧密的小型社群结构将被打破，脱离率会急剧攀升，因此其扩张上限非常有限。
关于 Jeremy Gaffin 提出的人们总在寻找“祭司阶层（priestly class）”，且这一角色近年来已从亿万富翁转移到网络发帖者（online posters）的观点，Cowen 持保留意见。他分析网络大 V 制造影响力的巅峰期可能早在五到十年前就已经过去；如今视频媒介迅速崛起，且 AI 生成内容本身也在不断挤占这些网络写手的生存空间。
### [23:21](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1401s) 大众创业的虚幻与职场人在 AI 时代的生存法则
在探讨就业前景时，Cowen 直言不讳地指出，绝大多数年轻人不可能也不应该成为创业者。社会上真正具有野心的人只占极小比例，大多数人更喜欢在大机构里工作，因为大机构薪水更稳定、风险更低，这一基本人性不会因为科技而改变。
针对那些身处大型组织、目睹 AI 变革却无力成为“AI 狂热者（AI maniacs）”去冒险创业的普通员工，Cowen 给出了务实建议：
- **主动拥抱前沿模型**：哪怕需要投入大量时间，也必须学会跟进并熟练操作最顶尖的模型，否则必然被淘汰。
- **所谓“休闲红利”尚未到来**：虽然行业整天大谈 AI 带来的空闲与效率红利，但现实往往要求绝大部分人加倍努力工作才能跟上步伐。
- **拒绝做被动的跟随者**：主持人引用 Joe Wiesenthal 的观点，认为 AI 极其擅长教学，因此“不存在学习曲线”，只要提问即可；但 Cowen 警示，如果永远只在后面追赶、被动等待模型指引，就无法建立卓越的竞争优势。既要了解技术战局的整体版图，又要吃透具体执行的实操细节。
### [26:41](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1601s) 赌博成瘾的蔓延与监管政策的现实困局
讨论随后转向蓬勃发展的博彩业。Cowen 坦言现在的赌博产业已经庞大到令人担忧的地步且极易成瘾，但他对“能否倒转时钟彻底解决”持悲观态度。针对是否应当“一键全面禁止体育博彩”的假设，Cowen 明确表示反对，直言历史已经证明强行禁止无法解决问题，当务之急是做出一份严谨的“成本效益分析（cost-benefit analysis）”，计算大众获得的娱乐效用与成瘾毁灭生活的社会成本之间的对比。
围绕博彩与烟草管控的异同，嘉宾与主持人展开了深层剖析：
- **心理机制与矛头转移**：赌徒即使输得倾家荡产，也很少仇视下注 App 或博彩公司（如 DraftKings），而是将愤怒转嫁到错失关键进球的运动员身上，在网络上进行网暴。
- **缺乏震撼性的健康危害指标**：反吸烟运动拥有“吸烟每年致死数百万人”这类杀伤力极强的健康数据，而博彩亏损在某种程度上被大众视作类似“花钱看3小时电影”的娱乐消费，缺少立竿见影的威慑性宣传口号。
- **负外部性的体验差异**：Cowen 补充道，控烟运动之所以取得实质成功，关键在于二手烟给周围不吸烟的人带来了直接的感官困扰与二手伤害（例如在飞机、电影院内闻到烟味）；相比之下，赌博的负面影响更为隐蔽，缺乏激起公众普遍反感的直观外部性痛点。
### [29:45](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1785s) 飞机禁烟的历史闲聊与手机博彩对比
对话从机上行为的比喻切入。主持人谈到，如果邻座乘客在飞机上玩手机博彩，自己完全无所谓，因为这根本不会干扰到他人；这与过去老式客机上允许吸烟的情况形成了鲜明对比——老飞机不仅在座椅扶手里内置烟灰缸，而且机舱后部还专门设有吸烟区，不吸烟的乘客不得不强忍二手烟的困扰。
大家顺势聊起了航空禁烟令的历史节点。有人依稀记得是在 80 年代末或 90 年代初全面禁止的，经核实确认：美国在 1988 年率先禁止了飞行时间在 2 小时以内的国内航班吸烟，随后在 1990 年将该禁令扩大至所有国内航班。嘉宾笑着调侃称，当年人们终于觉得“两个小时不抽烟还是能熬过去的”，这标志着公共出行环境的彻底转变。
---
### [30:35](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1835s) 就业数据评估：AI 时代下的劳动力市场真实状况
话题随后转入近期的宏观就业数据。主持人指出，全美整体失业率看似保持稳定，但增长动力似乎主要依赖医疗护理领域的招聘支撑，其他行业略显停滞，这印证了某种“脏活累活才招人”（messy job thesis）的现象。
```mermaid
flowchart LR
accTitle: 宏观劳动力市场当前现状
accDescr: 展现当前就业市场中青年失业率、移民因素与整体经济的基本面走势
subgraph MacroData["宏观基本面表现"]
direction TB
A["18-24 岁青年失业率"] --> B["回到大语言模型爆发前水平"]
C["移民数量减少"] --> D["新增非农数据表面放缓"]
end
subgraph EconomicOutlook["宏观经济走向"]
direction TB
E["劳动力市场温和降温"]
F["经济未见衰退拐点"]
E --> G["经济保持稳韧前行"]
F --> G
end
B -.-> G
D -.-> G
```
Tyler Cowen 对此给出了相对乐观的经济学解读。他特别强调，18 至 24 岁年轻群体的失业率目前已经回落到了大语言模型（LLM）大规模普及之前的水平，这直接反驳了“AI 正在摧毁初级就业市场”的恐慌论调。
虽然整体劳动力市场确实有所放缓，且受移民流入减少的影响，部分报告的表面数字显得不够亮眼，但在 Tyler 看来这属于正常周期调节。宏观经济并没有出现硬着陆或崩盘的迹象，短期内发生衰退的可能性很低。
---
### [31:47](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1907s) 结构性转型与地位洗牌：受教育阶层的焦虑 vs “AI 极客” 的崛起
面对“年轻人步入职场感知极差”的普遍舆论（即所谓职场氛围衰退），主持人询问这是否是因为科技大厂正在紧缩开支、削减了大量轻松光鲜的“笔记本电脑办公岗位”（laptop jobs）。Tyler 指出，当前正在发生的是剧烈的“行业结构性转移”（sectoral shifts）。
那些自视甚高、接受过良好精英教育的年轻人群体，对于新涌现出来的许多岗位要么看不上眼，要么根本不具备胜任的技能。由于这部分人在媒体上拥有极大的声量，他们的不满和抱怨被不成比例地放大了。
与此同时，社会正在发生深层次的阶层与声誉重组：
- **受教育精英的失落**：传统体面的白领上升通道正在变窄，引发了这部分精英青年的强烈焦虑。
- **“AI 极客”的跃升**：深度拥抱、驾驭 AI 工具的实践者们（AI maniacs）正迎来地位的飞跃式跃迁。
- **政治层面的震荡**：这种由技术驱动的社会地位大规模再分配（reallocation of status），将对现有政治体制带来巨大的消化压力。
在讨论斯坦福大学计算机专业应届生遭遇求职困难的现象时，主持人和 Tyler 均认同团队同事的评价——这在很大程度上是“能力与思维方式的问题”（skill issue）。对于拥有名校顶尖学历的人而言，若能及时转变惯性思维、主动创业或灵活选择赛道，机会依然充沛。
---
### [33:15](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1995s) AI 裁员的公关困境与工作认同的历史变迁
主持人和 Tyler 探讨了企业在 AI 转型期的人力策略。主持人批评不少企业借 AI 之名进行裁员是极度糟糕的公关策略。若企业自身拥有深厚的业务护城河，面对生产力工具爆发的契机，反而应当积极进取、大举进攻，例如 Cloudflare 首席执行官 Matthew Prince 逆势招募了上千名实习生。精通 AI 工具的年轻人加入具备壁垒的企业后，所能撬动的杠杆效应将呈几何级放大。相反，借 AI 裁员往往只是为了掩盖此前盲目扩张、过度招聘（hiring glut）而不得不进行的整顿。
Tyler 补充道，AI 核心原生企业自身的招聘需求一直非常旺盛。紧接着，主持人抛出了一个深层的社会学疑问：在人类历史上，是否曾有过多数人的自我认同与价值感不与“工作/雇佣”绑定的时期？Tyler 刚提到“中世纪”时代普通人以完全不同于现代的方式耕作生活，网络便出现了短暂卡顿，现场主持人打趣称他的网络“一下子掉回了 1988 年”。
访谈在轻松幽默的气氛中收尾。Tyler 幽默地给主持人留下一句建议：“下次做播客要加倍努力工作（work harder）”；主持人们则盛情邀请他下次到访加州线下演播室，承诺用极致讲究的手冲咖啡招待他，甚至尝试通过“施加同侪压力”拉他录制一场长达 4 小时的深度长谈。
## Notable quotes
> [00:06](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=6s) 我不喝咖啡。 — Tyler Cowen
> [00:06](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=6s) 这就是巅峰了，我不可能再尝到比这更好的了，因此无法被超越。 — Tyler Cowen
> [01:34](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=94s) 我希望并恳请特朗普政府放弃这场十字军东征，因为这根本不可能行得通。 — Tyler Cowen
> [01:34](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=94s) 当这些模型能在端侧设备上直接运行时，你又该怎么办？距离那一天还能有多久？既然如此，为什么一开始要走上封禁这条死路？ — Tyler Cowen
> [03:13](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=193s) 如果欧洲与中国爆发全面贸易战，欧洲必输无疑。 — Tyler Cowen
> [04:25](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=265s) 这两者在事实上是互补品。 — Tyler Cowen
> [06:16](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=376s) 如果马克·扎克伯格需要建数据中心，他完全可以直接给路易斯安那州州长打电话，谢天谢地这种沟通机制在我国切实有效。 — Tyler Cowen
> [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) 不要道歉，继续前行。 — Tyler Cowen
> [17:28](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1048s) Censure is the tax a man pays to the public for the privilege of being eminent — Tyler Cowen
> [33:15](https://www.youtube.com/watch?v=unxcoZbnH_Q&t=1995s) 下次做播客要加倍努力工作 — Tyler Cowen
## Mentioned
Tyler Cowen, 月之暗面 (Moonshot AI), Kimi k3, 特朗普政府, 英伟达 (NVIDIA), NVL72, 比亚迪 (BYD), 马克·扎克伯格 (Mark Zuckerberg), Thinking Machines, 凯拉·斯坎伦 (Kyla Scanlon), GPT-5, 《奥德赛》 (The Odyssey), 马特·达蒙 (Matt Damon), 赞达亚 (Zendaya), Collison 兄弟, Spotify, 乔纳森·斯威夫特 (Jonathan Swift), 马尔科姆·格拉德威尔 (Malcolm Gladwell), 史蒂芬·列维特 (Steven Levitt), 保罗·克鲁格曼 (Paul Krugman), Anthony Levandowski, Jeremy Gaffin, Joe Wiesenthal, DraftKings, Matthew Prince, Cloudflare, 斯坦福大学
## Video description
This is our full interview with Tyler Cowen.
We discussed why he believes banning Chinese open source AI is impossible, the rise of "AI maniacs," why Europe is quietly becoming an AI powerhouse, the future of jobs in an AI economy, why data centers are good for local communities, AI-generated religions, the gambling boom, economic "vibecession," and much more.
TBPN is made possible by:
Ramp - https://ramp.com
Public - https://public.com
Cisco - https://www.cisco.com
Console - https://www.console.com
CrowdStrike - https://www.crowdstrike.com
Figma - https://www.figma.com
MongoDB - https://www.mongodb.com
NYSE - https://www.nyse.com
Railway - https://railway.com
Shopify - https://www.shopify.com/
Codex - http://openAI.com/codex
Sign up for TBPN’s daily newsletter at TBPN.com
Follow TBPN:
https://TBPN.com
https://x.com/tbpn
https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231
https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235
https://www.youtube.com/@TBPNLive
[Source](https://www.youtube.com/watch?v=unxcoZbnH_Q)
Source quotes
[0:06] Tyler Cowen: 我不喝咖啡。
[0:06] Tyler Cowen: 这就是巅峰了，我不可能再尝到比这更好的了，因此无法被超越。
[1:34] Tyler Cowen: 我希望并恳请特朗普政府放弃这场十字军东征，因为这根本不可能行得通。
[1:34] Tyler Cowen: 当这些模型能在端侧设备上直接运行时，你又该怎么办？距离那一天还能有多久？既然如此，为什么一开始要走上封禁这条死路？
[3:13] Tyler Cowen: 如果欧洲与中国爆发全面贸易战，欧洲必输无疑。
[4:25] Tyler Cowen: 这两者在事实上是互补品。
[6:16] Tyler Cowen: 如果马克·扎克伯格需要建数据中心，他完全可以直接给路易斯安那州州长打电话，谢天谢地这种沟通机制在我国切实有效。
[17:28] Tyler Cowen: 不要道歉，继续前行。
[17:28] Tyler Cowen: Censure is the tax a man pays to the public for the privilege of being eminent
[33:15] Tyler Cowen: 下次做播客要加倍努力工作
Source quotes
[17:28] Tyler Cowen: don't apologize and keep on going
[17:28] Jonathan Swift: Censure is the tax a man pays to the public for the privilege of being eminent.
[22:51] Tyler Cowen: Follow AI. Learn how to work with the leading models.
Original transcript
[0:00] And without further ado, our first guest
[0:02] of the show is in the waiting room. We
[0:04] have Tyler Cowen, an American [music]
[0:06] economist,
[0:07] to talk to us about all the things. How
[0:08] are you doing, Tyler? Good to see you
[0:09] again.
[0:10] >> Hello. How are you? I don't drink
[0:11] coffee.
[0:12] >> I was about to ask.
[0:13] I have to ask every guest of the show.
[0:15] >> Are you going to start?
[0:17] >> I've only had coffee twice in my life,
[0:19] both times in rural Ethiopia in a coffee
[0:22] ceremony. And it was amazing, but I
[0:24] thought, this is the peak. I can't beat
[0:26] this.
[0:27] >> Is that because you'd become too
[0:28] powerful, or do you see it as a Faustian
[0:30] bargain?
[0:31] >> I see people getting addicted to it, and
[0:33] if they can't get their morning coffee,
[0:35] they have headaches. I don't want to be
[0:36] there.
[0:37] >> Yeah. It It does become a supply chain,
[0:39] though.
[0:40] >> How often are you somewhere where you
[0:42] can't possibly get a coffee in the
[0:45] morning?
[0:47] >> I don't know. You can't, but you have to
[0:48] go out and get it or go to some effort,
[0:51] or you're in a hotel room and you make a
[0:52] crummy coffee there, and then you don't
[0:54] really enjoy it. It's just simpler not
[0:57] to have that addiction, I think.
[0:59] >> What's very interesting is that uh
[1:01] coffee uh caffeine pills are incredibly
[1:05] economically efficient. It's like 1 cent
[1:08] for the equivalent of a cup of coffee if
[1:10] you buy them in bulk, and yet absolutely
[1:12] zero adoption. People love a cup of
[1:14] coffee even if it's a couple bucks. It's
[1:16] like a simple pleasure. It's a joy.
[1:18] People just can't stop paying for them.
[1:19] >> Do you do anything in the morning, any
[1:21] type of ritual that you feel like you
[1:22] have to do to be yourself?
[1:25] >> He's like Celsius.
[1:27] >> [laughter]
[1:27] >> No.
[1:28] I get high from mineral water, but if I
[1:30] don't have it, I'm fine. I'm not craving
[1:32] it.
[1:32] >> Not addicted. Okay. Okay. Well, uh let's
[1:35] talk about artificial intelligence. Uh
[1:38] how are you thinking about the latest
[1:40] debate around Kimmy K3 open-source AI?
[1:44] Where should we begin the discussion, uh
[1:47] and then I'm sure we can narrow down a
[1:49] bunch of different ways.
[1:51] >> Open source and is coming and is here,
[1:54] whether we like it or not. Of course, a
[1:56] lot of it will be from China. We should
[1:58] allow American competitors to proceed on
[2:01] an even footing.
[2:02] But basically, the attempt to outlaw it
[2:04] or ban it or use sanctions against it is
[2:06] going to fail miserably. Like, what
[2:08] about multinationals? What about
[2:10] companies whose affiliates use open
[2:12] source? What about all the major US
[2:14] companies that use open source Chinese
[2:16] products right now?
[2:18] Uh I hope I plead that the Trump
[2:20] administration gives up on this crusade.
[2:23] It cannot work.
[2:24] >> Okay. So, I want to push back on on it
[2:27] cannot be banned because I At first
[2:30] blush, I agree with you. Like, it is
[2:31] software. It is something that can just
[2:33] be downloaded. You can download the
[2:34] weights and just have it. Anything that
[2:36] that that's that's replicable infinitely
[2:39] uh should just proliferate across the
[2:41] world very quickly. Um but the fact that
[2:45] you need an NVL 72, you need a whole
[2:48] server rack, you need a equipment and
[2:50] infrastructure that acts as a point of
[2:53] leverage. I'm not saying that we
[2:54] shouldn't ban it. I'm just saying if you
[2:56] put me in charge of banning it, I think
[2:58] I would have an easier time banning open
[3:00] source AI than banning music
[3:03] distribution, for example.
[3:05] >> What will you do when it's on device?
[3:07] How long is that from now, right? Why go
[3:09] down the path at all?
[3:10] >> Yeah. Yeah. Yeah. Yeah. Yeah. That Yeah,
[3:12] that's a good point.
[3:12] >> Yeah, what what uh what lessons are
[3:15] there from the automotive industry for
[3:19] this moment? Um
[3:21] just just given, you know, the the use
[3:23] of of tariffs and how that those have
[3:26] worked and haven't worked and uh we were
[3:28] talking, you know, earlier about how
[3:31] Europe's approached the problem. You see
[3:33] BYDs all over Europe now.
[3:37] Uh but at the same time,
[3:39] many of the manufacturers have been hurt
[3:41] most by just losing the Chinese market,
[3:44] right? Or not losing it, but having
[3:46] declining sales there because of local
[3:48] alternatives.
[3:50] >> I think if Europe starts a major trade
[3:52] war with China, Europe will lose.
[3:55] This is a sad reflection on Europe and
[3:57] it's a sad reflection on Europe that
[3:59] they can't just say, "Oh, we'll make
[4:01] fewer cars. We'll do five or six other
[4:03] great things that we're readily
[4:04] positioned to shift into."
[4:06] They're not really there.
[4:08] Uh but simply putting up big tariffs
[4:11] isn't going to help them get there. It
[4:12] actually will postpone it. And those
[4:14] tariffs over time end up applying to the
[4:16] inputs for your other sectors and it
[4:18] makes them less competitive. So,
[4:20] Europe's really in a pickle. There's no
[4:22] good way out, but I don't favor the
[4:24] tariffs either.
[4:25] >> How do you think about uh
[4:28] the open-source models um
[4:31] in in terms of like the broader AI
[4:34] stack? Like are are is it important that
[4:37] America have some position in energy,
[4:41] data centers, uh chip manufacturing,
[4:44] networking? There's this whole stack and
[4:47] I've been surprised that we've been
[4:50] building as many data centers as we have
[4:52] in the United States given all the
[4:53] pushback. That feels like something that
[4:55] could have gone to at least other
[4:56] Western nations, uh energy-rich
[4:59] countries, but that's been fairly
[5:01] centralized. Um but there's there's
[5:04] clearly It feels like there's a
[5:05] potential slippery slope for Oh, well,
[5:07] if China's handling the open-source
[5:09] models, then maybe they should have the
[5:10] data centers. Maybe they should also do
[5:12] the power generation over there. It's
[5:13] dirty. We don't like having data centers
[5:15] here. Put them over there. And then at
[5:16] some point it feels like you could lose
[5:18] leverage and but I don't know how risky
[5:20] that actually is.
[5:22] >> I live right next to all the data
[5:23] centers in Northern Virginia. It's
[5:25] great. They pay for half of the taxes of
[5:27] their county. The energy, the envy, all
[5:30] that, I'm completely for it. The chips,
[5:32] bring it on. But at the end of the day,
[5:34] people are afraid that the American
[5:36] models can be yanked from them.
[5:38] >> Sure.
[5:38] >> That's partly our fault. And if there's
[5:40] open-source as something in the
[5:42] background, uh people are more willing
[5:45] to lock into our systems to a partial
[5:47] extent. So, the two are complements in
[5:49] fact. And again, there's all these major
[5:51] American companies that don't typically
[5:53] advertise it, but they use a lot of
[5:55] Chinese open source right now.
[5:57] >> Yeah.
[5:57] >> You just going to pull that out of their
[5:59] guts?
[6:00] >> Yeah. Yeah. I mean, both enterprises
[6:02] actually deploying these models for its
[6:04] cost savings and also I mean, we saw
[6:06] with Thinking Machines they they used
[6:07] the Chinese open source model to
[6:09] bootstrap a step in the process of
[6:11] building their latest product. And so,
[6:13] it's a whole supply chain.
[6:15] Where do you think the data center
[6:17] pushback comes from? If you are there
[6:19] and have a positive experience, it's
[6:21] polling very poorly. What Is it all
[6:25] Is it all fear around job displacement?
[6:27] Is it anything particularly on the
[6:29] ground? Like where where did this come
[6:31] from?
[6:32] >> I think right now people are just
[6:33] negative about many, many things.
[6:36] They're negative about their country.
[6:38] They're negative about uh
[6:40] social institutions except the military.
[6:43] They're very low levels of trust. They
[6:45] were told during COVID times, during the
[6:47] great financial crisis, everything's
[6:49] okay when it was not. So, you can't
[6:51] completely blame them. But look, as you
[6:54] know, this country has 50 states. And if
[6:56] Mark Zuckerberg needs to pick up the
[6:58] phone and call the governor of
[6:59] Louisiana, thank goodness that can work
[7:01] here.
[7:02] And I think we're going to do it.
[7:05] >> What is it about this We're in this like
[7:07] counterintuitive vibe session moment.
[7:09] Kyla Scanlon sort of
[7:11] written about it, coined the phrase.
[7:13] This idea that
[7:14] when
[7:16] when you pull people on how are your
[7:17] neighbors doing economically or
[7:19] financially, people say, "They're doing
[7:20] fine, but I'm doing terribly." Or the
[7:23] economy will grow, the job market will
[7:25] be stable, but people will report the
[7:27] consumer confidence is very low.
[7:29] I'm wondering if you've reflected on
[7:30] that more generally. And then I do have
[7:32] a follow-up question. But
[7:34] what is the state of the vibe session?
[7:35] Is it just continual? Is it permanent?
[7:38] Is it getting worse?
[7:40] >> I call it negative emotional contagion.
[7:43] And you see it in many other areas of
[7:44] world history. It feeds on itself. I'm
[7:47] not sure we know what starts it. Maybe
[7:49] it is some negative events. But once it
[7:51] gets going, it's hard to reverse.
[7:54] Sometimes you need actually a war or
[7:55] some kind of catastrophe. I hope we do
[7:57] not.
[7:59] But if you look at wealth accumulation,
[8:00] the job market, stock prices, real
[8:03] wages, many other indices, they're doing
[8:06] like okay to fine. They're just not
[8:08] doing terribly.
[8:09] >> Yeah. No, no, I completely agree. And so
[8:12] I had this idea, I don't know if you'll
[8:14] agree, but
[8:16] maybe the fact that we're
[8:19] in this vibe session and the vibes are
[8:21] so bad and there's so much negative
[8:23] emotion around the economy that if we
[8:26] actually were to go through a brief
[8:29] recession or a market correction,
[8:32] we might be a more resilient country
[8:35] sort of ironically because of that
[8:36] because everyone would look around and
[8:38] be like well, I already thought it was
[8:40] bad. It got a little worse, but I'm
[8:42] ready to reinvest. I'm maybe this is
[8:44] what I've been waiting for. Finally, I'm
[8:47] justified in thinking things are bad and
[8:49] so I can get out of this and do
[8:52] whatever's necessary because I am I'm
[8:55] there's no longer a cognitive dissonance
[8:57] between the what the market is doing and
[9:00] how I feel.
[9:01] >> One hopes. Yes, it could be a bullish
[9:03] signal. But then again, if something
[9:05] terrible does happen, everyone says I
[9:07] told you so and maybe we get stuck in
[9:09] the mire.
[9:10] >> Yeah.
[9:11] >> Jordy.
[9:11] >> About a year ago,
[9:13] you had a blog post if I remember
[9:15] correctly, one of many of course, that
[9:18] that was something to the effect of like
[9:20] AI is real and it's amazing, but the
[9:22] economy is still going to just sort of
[9:24] generally continue growing as it's
[9:26] grown, maybe a little bit faster,
[9:28] hopefully.
[9:30] Are you seeing anything that could
[9:32] update your world view on that because
[9:35] so far I feel like you've been pretty
[9:37] vindicated.
[9:37] >> Deserve a victory lap.
[9:39] >> Well, I'm seeing a lot of confirmation.
[9:41] Through so many people that will say to
[9:43] me or say in chat groups or on Twitter,
[9:45] my goodness, if we had Fable 5, I would
[9:47] have thought the world would be totally
[9:49] different and it's not.
[9:50] >> Yeah.
[9:51] >> And that's my view, but it look, you go
[9:53] back to 1995 through 1998, we got half a
[9:57] percentage point of extra growth out of
[9:59] those advances. They to me seem less
[10:01] impressive than AI. So, I think over
[10:03] time with some slowness, we'll get that
[10:06] and maybe a bit more, which is great.
[10:09] >> Interesting.
[10:10] Uh have you seen The Odyssey?
[10:13] >> I have seen The Odyssey.
[10:14] >> Can you give us your review?
[10:17] >> Well, look, compared to other movies,
[10:19] it's fantastic and it's dreamy and it's
[10:22] impressive, but mostly I was
[10:24] disappointed. I thought the dialogue was
[10:26] lame. The casting, I don't mind the
[10:28] cross-ethnic casting at all. I just
[10:30] didn't think they were the right people.
[10:32] Matt Damon is not Odysseus. He's just
[10:35] not impressive enough physically. The
[10:38] woman who played Calypso just seemed
[10:40] like a suburban housewife. I love
[10:42] Zendaya, but she didn't have any real
[10:44] role. So, they screwed up many things.
[10:48] It didn't quite cohere.
[10:50] Uh I'm very glad I saw it. Everyone
[10:52] should see it. It's worth the 3 hours,
[10:54] but relative to expectations, I suppose
[10:56] a disappointment.
[10:57] >> Do you think it's Do you think
[10:59] classicists are satisfied with it? Do
[11:01] you think it's good for the field?
[11:04] >> They're delighted to have any attention
[11:06] at all
[11:06] >> [laughter]
[11:07] >> and the different translations of The
[11:08] Odyssey are now on the best seller
[11:10] lists. So, what's their alternative,
[11:13] right? They shouldn't complain too
[11:15] loudly.
[11:16] >> Uh do you think technology is making
[11:20] soccer worse?
[11:23] >> I don't watch soccer,
[11:25] but I think technology in some ways has
[11:28] made basketball worse and I wouldn't be
[11:30] surprised if the same were true of
[11:32] soccer. That it homogenizes things and
[11:35] everyone chases after the same targets,
[11:37] the same players, teams have less
[11:39] individual style. Uh but to soccer I
[11:42] cannot speak.
[11:43] >> Hm.
[11:44] Fair.
[11:44] >> Um
[11:45] what about AI maniacs? Take me through
[11:48] what defines an AI maniac, why it's
[11:50] important, and I have some uh some
[11:53] pushback and follow-up questions.
[11:55] >> I spoke to two teenagers this morning.
[11:58] One was 13, one is 17. Abu Dhabi,
[12:02] Finland. They're AI maniacs. They spend
[12:04] their spare time mastering AI. They know
[12:07] way more than their so-called teachers.
[12:10] They're starting companies or building
[12:12] models. They realize this is the future.
[12:15] You're going to have a large number of
[12:17] companies with quite a
[12:19] very small number of employees, but
[12:21] pretty high revenue. And I think AI
[12:23] maniacs will compete in medicine, law,
[12:26] business, consulting, banking, wherever
[12:28] it is we let them compete. So this to me
[12:30] is the future. Like if you're asking,
[12:32] will the UK recover?
[12:34] I would rephrase it. Is the UK going to
[12:36] encourage its AI maniacs?
[12:38] >> How can they encourage these AI maniacs?
[12:41] It feels like AI naturally lowers the
[12:43] barrier to entrepreneurship, and so uh
[12:46] there's not much encouragement that's
[12:48] necessary. I mean, as long as the models
[12:50] are available and affordable for young
[12:52] people, they will do things with them.
[12:55] >> Young people should be allowed to rise
[12:57] to positions of authority. They should
[12:59] be respected. They shouldn't be looked
[13:01] [clears throat] down upon. They should
[13:03] have the chance to buy computer, get
[13:04] grants. It's not automatic.
[13:06] >> Hm.
[13:06] >> So you look at the United States, we do
[13:09] great on all these metrics. Estonia is a
[13:12] country that does pretty well on these
[13:13] metrics. A lot of parts of the world are
[13:16] too small, too conservative.
[13:18] And we'll see where different countries
[13:20] are going to fall on this spectrum. But
[13:21] they're not all going to do it
[13:22] automatically, just like not all
[13:24] countries have made great rock and roll.
[13:27] Though, the technology is pretty
[13:28] straightforward.
[13:29] >> Yeah. Uh yeah, Estonia is famous for um
[13:33] uh
[13:35] major government investing in in
[13:37] high-speed bandwidth. Is that correct?
[13:39] Is that the Is that the rough story of
[13:40] Estonia?
[13:41] >> Yes. Yes.
[13:41] >> Um and but I'm interested in what that
[13:45] would look like for AI in America or the
[13:48] UK. Like it feels like many of the
[13:51] models or application layer companies,
[13:54] they have sort of a free tier that gets
[13:57] someone potentially from zero to close
[14:00] to AI maniac and then at some point they
[14:03] can go and find a customer to help pay
[14:05] for their compute bills. But is there
[14:07] more the government should be doing?
[14:09] >> Well, as you know, with venture capital,
[14:11] there's a valley of death. I don't know
[14:13] that the government should be making
[14:14] grants to 16-year-olds, but someone
[14:16] should. I'm doing it. Other people
[14:18] should as well. So, this is really the
[14:21] province of youth. There's nowhere you
[14:23] can go to learn this stuff other than
[14:25] say some of the major AI centers and
[14:27] companies. So, you've got to teach
[14:29] yourself. And is there the support in
[14:31] your environment for that? We will see.
[14:34] >> Give uh give us an update on how you're
[14:37] thinking about grants. Are you going
[14:39] younger, broader, smaller, different
[14:42] partners? I mean, uh there's been some
[14:45] folks who have been able to do compute
[14:46] grants and then they sort of cap they
[14:48] sort of the dollars go further because
[14:50] of the implicit margin.
[14:53] >> With AI maniacs, I'm going younger and
[14:56] weirder. And I've also noticed Europe is
[14:58] doing pretty well on the AI maniacs
[15:01] metric. We'll see how it develops as the
[15:03] costs rise.
[15:04] >> Yeah.
[15:05] >> Uh but I've been pleasantly surprised
[15:07] and I'm now more optimistic about Europe
[15:10] than I was 6 months ago.
[15:11] >> Mhm.
[15:12] Uh how does that uh like
[15:15] where did that come from? That feels
[15:16] like a come from behind story for
[15:18] Europe. I feel like the for the For the
[15:20] time the the brightest technologists in
[15:23] Europe have gone to America to start
[15:25] their companies. The the Collisons,
[15:27] famous example, there's many others.
[15:29] Uh with the exception of like the Swedes
[15:30] and Spotify, I suppose. But in general,
[15:33] it feels like many people have come. Are
[15:35] they staying more frequently? Like
[15:36] what's happening?
[15:38] >> Well, these kids may yet go to America.
[15:40] They're too young to be going now.
[15:43] Uh maybe it's that
[15:45] you know, chatbots didn't interest the
[15:46] Europeans that much. They seem to like
[15:48] compete with culture.
[15:49] >> Yeah.
[15:50] >> Uh agents are a chance to do something.
[15:53] Now, American kids have had
[15:54] opportunities to do things for a long
[15:56] time.
[15:57] >> Yeah.
[15:57] >> But for the European kids, this is
[15:58] something new. So, I think at the
[16:00] margin, you'll see more extra activity
[16:02] there.
[16:03] >> Yeah.
[16:03] >> And a lot of new companies from quite
[16:05] young people coming out of all of
[16:07] Europe, including Eastern Europe.
[16:09] >> Do you think young people are better at
[16:11] avoiding the fake work trap that happens
[16:15] with some AI users? You know, the person
[16:17] that just builds an automated workflow
[16:20] to, you know, check, summarize every
[16:22] meeting, and then they never look at the
[16:24] notes, and it's just endless token
[16:26] consumption in service of like creating
[16:29] the ultimate second brain that they
[16:30] never actually refer to. And it's all
[16:33] it's ultimately a lot of make-work, but
[16:35] it feels very productive, sort of the
[16:37] tool-shaped shaped objects formulation.
[16:40] >> I think the young people are more
[16:42] playful with it. They take more chances.
[16:44] In the aggregate, they might even do
[16:46] worse. But in terms of coming up with
[16:48] the new things that really matter, I
[16:49] think they're going to win.
[16:51] >> Mhm.
[16:51] Uh what what do you think about the the
[16:54] risk of of the one-shotting from AI?
[16:56] People that go so deep in it, they lose
[16:59] they lose their grip on reality.
[17:03] >> It happens with everything, you know.
[17:05] When I was a kid, people one-shotted on
[17:07] baseball. Can you imagine that? Or like
[17:09] drugs.
[17:10] >> Yeah.
[17:11] >> Uh AI's one of the better things to be
[17:13] one-shotted from, I would say.
[17:15] >> No, I I I was laughing about it. know
[17:16] some guys who have been basically one
[17:18] shot by golf. And they and all day,
[17:21] every day. They're like reading about
[17:23] golf, studying golf, trying to improve
[17:24] their golf game. And it's the same thing
[17:26] as AI. Yeah.
[17:28] >> What uh
[17:29] if a billionaire came to you and said,
[17:32] "Tyler, it seems that my type is very
[17:35] unpopular right now.
[17:37] What should we do to to try to uh shift
[17:42] public opinion? Uh right now there's a I
[17:46] think a $80 million campaign being spent
[17:48] to
[17:49] uh uh block the the the billionaire tax
[17:53] here in California. But it seems like
[17:55] there are other things more broadly that
[17:58] that billionaires could do to
[18:00] uh shift public perception. But what do
[18:02] you think?
[18:03] >> I would say don't apologize and keep on
[18:06] going. I hope the billionaire tax fails.
[18:09] It will wreck California if it passes.
[18:11] It's already harmed California.
[18:13] Even a lot of the Democratic Party
[18:15] politicians don't support it.
[18:18] Uh
[18:19] I think again there's just this negative
[18:21] sentiment about many many things
[18:23] including billionaires. You might say
[18:25] some of them deserve it. Fine. Like with
[18:26] all humans.
[18:28] But at the end of the day I'm reminded
[18:30] of the old saying from Jonathan Swift
[18:31] and he said, "Censure is the tax a man
[18:35] pays to the public for the privilege of
[18:37] being eminent." Unquote.
[18:40] >> Mhm.
[18:41] >> If you succeed, people will hate you.
[18:44] >> Okay, but
[18:45] >> You saw this you know, wasn't it Malcolm
[18:46] Gladwell, Steve Levitt, Paul Krugman?
[18:49] Whoever did well, got hated.
[18:51] >> Oh, yeah. That's right.
[18:53] >> At the same time, would uh
[18:56] building some massive, you know, parks
[19:00] and or
[19:02] you know, any any type of of public work
[19:04] >> Yes, I I think
[19:06] >> be be worthwhile?
[19:07] >> question is is there is there some sort
[19:09] of like
[19:10] >> Like a
[19:11] >> greater leverage from having like the
[19:13] library that you walk by versus like 1%
[19:18] lower levels of malaria that's like this
[19:20] abstract thing that you don't see. Like
[19:23] there's something about building like a
[19:25] monument as a wealthy person as opposed
[19:28] to having an abstract diffuse impact on
[19:31] humanity even if the reduction
[19:33] >> Maybe the modern version of a park is
[19:35] just like owning an NFL team.
[19:37] >> Uh it doesn't really feel the same
[19:39] >> The Sultan of Brunei just built a
[19:41] monument.
[19:42] >> True.
[19:43] >> love him for it?
[19:44] Now I'm all for today's billionaires
[19:47] doing more to support art, music,
[19:49] libraries, whatever we all might come up
[19:51] with, but I don't think it's going to
[19:53] make them loved.
[19:54] >> Yeah.
[19:55] Yeah, maybe not. Uh what about religion
[19:58] and AI? There were Anthony Lewandowski
[20:00] was working on an AI religion way before
[20:03] it was cool. I think 2017 he was
[20:06] thinking about this. Um
[20:07] where do you it feels like we're maybe
[20:10] behind the curve on
[20:13] a true AI-based religion. It feels like
[20:16] based on the capabilities of the models
[20:17] you could have something happening.
[20:19] There's little pockets of people that
[20:21] are obsessed with certain models, but uh
[20:23] I haven't seen anything formally being
[20:25] built out. What what where do you think
[20:27] that goes?
[20:28] >> I think it'll be more like sects than
[20:30] new religions. Mhm. So you could write
[20:32] your own version of Hindu sacred books
[20:35] or your own version of Shiite Islam with
[20:37] your own, you know, slate of who the
[20:39] true Imams are or should be.
[20:41] And we're just going to see
[20:42] proliferation. Like for a while in
[20:45] history we had a proliferation of
[20:47] different versions of the LDS Church.
[20:50] >> Mhm.
[20:50] >> Different kinds of Mormons. A lot Some
[20:52] are still around.
[20:53] >> Yeah.
[20:53] >> And using an agent you'll be able to
[20:55] whip up your own religion, I don't know,
[20:57] in less than a day
[20:58] with rituals, sacred books, practices.
[21:01] Most won't catch on, but a few will.
[21:04] They'll have a few hundred, a few
[21:05] thousand adherents. That may sound
[21:07] trivial, but if you know, half of 1% of
[21:10] the population does this and some of
[21:12] these get some number of adherents, over
[21:15] time the whole religious landscape looks
[21:17] quite different. I think it will be more
[21:18] diverse and feel weirder and be less
[21:21] centralized in religious authorities.
[21:24] >> Mhm.
[21:25] Uh
[21:26] where do you think the Amish will be in
[21:29] 200 years? I saw someone try and use one
[21:32] of these AI models to project out
[21:34] project out the human population in and
[21:37] in 20
[21:39] uh 2200
[21:40] and uh the prediction came back as like
[21:43] America will be 34% Amish. And I think
[21:45] it was just extrapolating, you know,
[21:47] linear trends of their birth rates
[21:49] versus everyone else. But uh
[21:51] it feels like the idea of the Amish
[21:54] lifestyle is actually becoming in vogue
[21:57] in almost popular culture with uh don't
[22:00] use screens as much, try and put uh
[22:03] limit your Instagram usage, you need to
[22:05] log off and touch grass. All of these
[22:08] memes are sort of it it uh
[22:10] you know, the the potential gateway drug
[22:13] to the Amish lifestyle.
[22:15] >> I visited the Amish many times. I admire
[22:19] what they've done, but it's possible
[22:21] because the surrounding society cross
[22:23] subsidizes them.
[22:25] >> Oh.
[22:25] >> So, you can't have a third of America
[22:27] being Amish. And I think they keep
[22:29] people in the fold because it is pretty
[22:31] small and tightly knit.
[22:33] And were there, say, three or four
[22:35] million Amish, which we're very far
[22:37] from, the rate of leaving would be much
[22:39] higher. So, I think it's bounded on the
[22:41] upside. But I'm glad they're there.
[22:43] They've done a great job. Their
[22:44] communities are quite interesting and
[22:46] pretty prosperous. They have a lot of
[22:48] kids. Good for them.
[22:51] >> Uh do you agree with Jeremy Gaffon that
[22:54] people are constantly looking for a
[22:55] priestly class and in recent years it's
[22:58] moved from the billionaire class to the
[23:00] online poster?
[23:03] >> I don't know. Maybe online posters
[23:06] peaked 5 or 10 years ago.
[23:08] And now there's a lot more video and
[23:10] just the AI itself competes against the
[23:13] online posters.
[23:14] >> Sure.
[23:14] >> Uh I liked his dialogue on that topic,
[23:17] but I don't exactly agree with what he
[23:19] said.
[23:20] >> Sure. Sure.
[23:21] Uh
[23:22] I I'm interested to know what
[23:26] level of entrepreneurship you think is
[23:29] is reachable. Can Can every young person
[23:32] become an entrepreneur? Or are there
[23:34] some people who
[23:36] are more interested in just finding a
[23:38] stable job and then they need to figure
[23:41] out how to navigate changing jobs in the
[23:43] future.
[23:45] >> Most people cannot become entrepreneurs.
[23:47] They're not ambitious. They should not
[23:49] be ambitious. You could argue the
[23:51] numbers, but it's a small percentage.
[23:54] And other people want to work for them.
[23:56] >> Mhm.
[23:56] >> So people like working in large
[23:58] institutions. Large institutions pay
[24:01] better. It's lower risk. I don't think
[24:03] that feature of the world's going to
[24:04] change.
[24:05] >> So advice for someone who's in an
[24:09] organization,
[24:11] they're seeing things change because of
[24:13] artificial intelligence.
[24:15] Maybe they're not set up to be an AI
[24:17] maniac, go out and start a company, but
[24:20] should they be trying to find a role
[24:22] that is messier in their organization?
[24:26] >> Absolutely. Follow AI. Learn how to work
[24:28] with the leading models. It's a big time
[24:31] investment cuz everything changes so
[24:32] quickly as you know.
[24:34] >> Yeah.
[24:34] >> But if you don't, you're left behind.
[24:36] And for all this talk of the leisure
[24:37] dividend from AI, right now we're still
[24:40] in the setting where most people have to
[24:42] work some amount harder.
[24:43] >> Yeah.
[24:44] >> change, but it's not about to change.
[24:46] Time to work. Get to work.
[24:48] >> Do you Do you Do you Do you Do you
[24:49] >> Doesn't it exhaust you how many new and
[24:51] different things happen in the world of
[24:52] AI? And you've got to stay on top of
[24:54] them.
[24:54] >> Yeah. Yeah, it is a lot.
[24:56] >> Where's your AI leisure dividend?
[24:57] >> It uh
[24:59] >> it is a lot. a lot, but at the same
[25:01] time, I don't know, learning learning
[25:03] about AI I see as different as actually
[25:06] learning AI. Like there are a lot of
[25:08] there are a lot of facts around the
[25:10] there are there are six AI tigers in
[25:13] China and like I just need to memorize
[25:15] what those six companies are, their
[25:17] histories if I want to understand the
[25:18] industry.
[25:19] >> follow it like a sport and not learn
[25:22] really anything at all about how what
[25:24] what the models are
[25:25] >> But that's different than using the
[25:26] models, which the models
[25:29] are so good it feels like it's it's less
[25:32] of a learning curve and more of just the
[25:33] volition to just go and do it because
[25:36] even when someone says, "Oh, well, I
[25:39] haven't tried the the latest coding
[25:41] models." And I'm like, "Oh, well, they
[25:43] can do a lot more." Like at the very
[25:44] basic, like if you wanted to ask ChatGPT
[25:47] for one image, you can now ask Codex for
[25:49] 100 images and it can just go and do it
[25:51] in a for loop and it can write a little
[25:53] code for you and do whatever to to get
[25:56] you your answer that's more robust and
[25:58] hydrated.
[25:59] But also you can just ask ChatGPT, "How
[26:02] do I do that?" and it will tell you how
[26:04] to do it. And so the learning curve this
[26:06] is a Joe Wiesenthal opinion, but he he
[26:09] says that like there there is no
[26:11] learning curve because the AI is good at
[26:13] teaching you how to use it.
[26:15] But there is something there where you
[26:17] talk to people all the time that are
[26:18] like, "Ah, I haven't tried AI yet. I got
[26:19] to check that out."
[26:22] >> It can tell you what to do, but if
[26:24] you're always chasing it, waiting for it
[26:26] to tell you what to do, you're not on
[26:28] top of it. It's not really the way to
[26:31] excel in the area. So I think you do
[26:33] have to stay on top of both like the
[26:36] baseball team side of it and the how do
[26:38] you do this side of it. Again, it's
[26:40] hard.
[26:41] >> How big can gambling get?
[26:45] >> Oh, it's already too big. I don't know
[26:48] how we pull it back in.
[26:50] Um
[26:51] it's a problem. Now people must enjoy
[26:54] it, the people who don't wreck their
[26:55] lives with it, or it wouldn't be that
[26:57] addictive. But, it is fairly addictive
[26:59] for many people.
[27:01] And I don't see how we turn the clock
[27:02] back.
[27:03] >> The thing that's telling is
[27:05] even people that lose a lot of money
[27:07] gambling, I've noticed that they don't
[27:10] hate their like bookie or the app that
[27:13] they use to lose money.
[27:15] >> Yeah.
[27:16] >> They seem to have like gen-
[27:19] you know, maybe maybe it's like a toxic
[27:21] relationship, right? Where where it's
[27:23] >> A lot of times people will hate like the
[27:25] player that missed the field goal, and
[27:27] they will be cyber bullying the
[27:30] particular athlete that lost them the
[27:32] money, and they won't be mad at the
[27:34] player.
[27:34] >> manager at DraftKings or whatever.
[27:35] >> Yeah. I mean, isn't
[27:37] >> But, how would what like, you know,
[27:39] you if you have a big red button that
[27:41] the ban, you know, sports betting or or
[27:44] gambling, do you do you hit the button?
[27:46] Like, what what would be your practical
[27:48] solution for reining it in?
[27:51] >> I wouldn't hit the button. Banning it
[27:53] would not rein it in. We know that
[27:55] already.
[27:56] Uh you hope norms change. It's a bad
[27:59] thing to be one-shotted by. That said,
[28:02] I'd like to see a cost-benefit analysis.
[28:04] All the people who enjoy it, and add
[28:06] them up, and see how that compares to
[28:08] the costs. No one wants to do that for
[28:10] obvious reasons.
[28:12] Uh just say no.
[28:13] >> it's but it but but
[28:16] right now
[28:17] >> don't do it.
[28:18] >> [laughter]
[28:18] >> Yeah, I guess I guess
[28:21] I I was watching uh the the latest uh
[28:24] UFC event, and almost every single ad
[28:28] is a gambling ad. Certainly reining in
[28:32] advertising for gambling would probably
[28:35] >> We have a playbook from the tobacco uh
[28:37] >> Yeah, and then I think it's in like
[28:39] Australia.
[28:40] >> You can restrict You can add taxes on a
[28:42] state and local level, as well as
[28:44] federal taxes, and you can sort of try
[28:46] and internalize the negative
[28:48] externality.
[28:49] >> was that law that said you could only
[28:50] write off like 90% of your losses, which
[28:53] which for you know, which is really bad
[28:56] if you're you know, a sports book or
[28:58] >> is is that I don't think the the impact
[29:01] of gambling the negative impact impact
[29:03] has been fully quanti-
[29:05] quantified in sort of a sound bite yet.
[29:08] Like with cigarettes it's like
[29:10] cigarettes kill millions of people every
[29:12] year. 50% of smokers who smoke a pack a
[29:15] day will die from cigarettes or a
[29:16] cigarette related to illness. There
[29:18] isn't that for the gambling industry.
[29:20] It's just like a lot of people lose
[29:21] money, but you technically lose money
[29:23] when you go see the Odyssey. You just
[29:25] get the enjoyment of a 3-hour film. You
[29:27] lost money and people get the
[29:29] entertainment of gambling and so that
[29:30] that that's the same trade. It doesn't
[29:32] hit as hard as like you will die if you
[29:34] smoke cigarettes.
[29:36] >> Tyler?
[29:36] >> the campaign against smoking worked
[29:38] because people like me hated it when
[29:40] someone next to them would light up or
[29:42] on a plane or in a movie theater.
[29:45] If someone if my next to next to me is
[29:46] on their
[29:48] phone
[29:48] >> Wait, really?
[29:49] >> on the airplane, I don't care.
[29:52] >> No, you used to Yeah, yeah,
[29:53] if you're on old commercial planes, they
[29:55] would have a cigarette
[29:57] an ashtray in the armrest.
[30:01] >> I suffered under that for years.
[30:03] >> Yeah.
[30:03] >> You try to not sit next to the smoking
[30:06] section.
[30:06] >> Yeah, it's the same thing.
[30:07] >> if there's like a a no gambling section
[30:09] on the airplane, it's ridiculous.
[30:11] >> Yeah.
[30:12] >> The guy next to me could be gambling on
[30:13] his phone.
[30:14] >> Yeah, you wouldn't even know.
[30:15] >> I don't know who Yeah, exactly. Let it
[30:17] be.
[30:18] >> Yeah.
[30:18] >> That is shocking to me.
[30:20] >> I like that you just learned that. When
[30:21] did When did smoking on planes end? In
[30:23] the '80s? I feel like it was not
[30:26] really around. Maybe like early '90s at
[30:29] the very very latest. Uh
[30:31] >> Yeah, I recall celebrating, but I don't
[30:33] know the year.
[30:34] >> Yeah.
[30:35] Um anyway, uh what what what have you
[30:38] been reading into the jobs numbers that
[30:41] have been coming out. Like the overall
[30:44] unemployment rate seems to be holding
[30:45] steady. It seems like we're buoyed by
[30:49] health care hiring. Yes.
[30:51] >> Smoking was banned in 1988 on domestic
[30:54] flights lasting 2 hours or less.
[30:56] >> Okay. You're saying you can go 2 hours
[30:58] without a smoke.
[30:59] >> Okay, yeah. [laughter]
[31:00] >> We know you can. And then eventually
[31:02] they expanded it to all domestic flights
[31:04] in 1990.
[31:05] >> Yeah, 1990. There you go.
[31:07] Uh jobs data, hiring, it feels like
[31:10] there's sort of stagnation everywhere
[31:12] except health care, which is sort of the
[31:14] messy job thesis playing out. But what
[31:16] else is going on in the hiring market
[31:18] that you've noticed from the last couple
[31:20] releases?
[31:21] >> Well, the unemployment rate for 18 to
[31:23] 24-year-olds, it's back to where it was
[31:26] before we had good large language
[31:27] models. So AI is not destroying that
[31:30] market. It seems we're not going to have
[31:32] a recession anytime soon. That's good
[31:34] news. Labor market has slowed down quite
[31:36] a bit, but that's to be expected. And
[31:39] some of the reports sound less
[31:40] impressive because we have fewer
[31:42] immigrants.
[31:43] But overall, I've been pretty happy. The
[31:45] economy hasn't crashed. It's kept on
[31:46] going.
[31:47] >> Yeah. Yeah, no, that makes sense. Uh
[31:49] then what are you attributing the like
[31:52] the vibes around like young people
[31:55] entering the workforce? Everyone says
[31:56] it's so bad. Is that just the really
[32:00] cushy tech jobs are sort of a little bit
[32:02] belt-tightening there? They're not
[32:03] hiring as much, and so there aren't as
[32:05] many of these like really nice laptop
[32:07] jobs? Or what are you
[32:09] What do you think's going on there?
[32:11] >> I think there are big sectoral shifts
[32:12] going on, and the new jobs being
[32:14] created, people who think of themselves
[32:17] as educated and high-class may not want
[32:19] or may not even be good at, and they
[32:21] complain, and they're vocal, and they
[32:23] have the ear of the media. It is a
[32:25] problem. I wouldn't deny that. But in
[32:28] the aggregate, a lot of other people are
[32:29] doing better, and the AI maniacs are
[32:32] about to take a huge leap upwards, and
[32:34] there'll will this massive reallocation
[32:36] of status. Yeah.
[32:37] >> It'll be very hard for our politics to
[32:39] handle that.
[32:40] >> Yeah, I saw Oh, that's interesting. I
[32:41] won't
[32:42] uh
[32:43] Well, well. Um uh yeah, I I I saw some
[32:46] stats around like uh uh
[32:48] new grad computer scientists from
[32:50] Stanford uh not getting as many job
[32:53] placements and uh Tyler on our team was
[32:56] saying like, "That's a skill issue. If
[32:58] you're if you're a Stanford graduate
[33:00] with a computer science degree, like you
[33:01] can get a job or you can start a
[33:03] company. You can do something, but you
[33:05] might need to think a little bit
[33:07] differently because the track might be
[33:08] slightly different." Um but I don't know
[33:10] if you agree with that or not.
[33:12] >> I agree with that.
[33:13] >> Yeah. Um and it also just feels like
[33:16] that one of my
[33:18] latest uh
[33:19] like thesis is uh
[33:22] that a lot of these CEOs who are doing
[33:23] like AI-based layoffs, the optics are
[33:27] just terrible and if they actually have
[33:28] a good business, they should be going on
[33:30] the offensive. We we we we we love that
[33:32] Matthew Prince hired, I think like a
[33:34] thousand interns or something like that.
[33:36] And it feels like young people who
[33:38] understand AI coming into an
[33:40] organization that has some sort of moat,
[33:42] some sort of durability, some business,
[33:44] like the leverage is just going up. So,
[33:46] you should be hiring more instead of
[33:48] doing layoffs if you actually understand
[33:51] what's at stake and how big your
[33:52] business can be. But if you just went
[33:54] through a hiring glut, maybe you do need
[33:55] to uh sort of retool.
[33:58] >> The AI companies themselves are hiring
[34:00] plenty, as you know, right?
[34:01] >> Yeah. Totally.
[34:03] >> Uh can you recall a point throughout
[34:06] history where
[34:08] the majority of people uh
[34:11] uh uh
[34:12] their their let's say their their sense
[34:14] of like self-worth and identity was not
[34:17] tied to their employment
[34:19] or their job?
[34:21] >> Hm.
[34:22] >> Because that's the big that's a big
[34:23] >> separate issue, but obviously that's a
[34:26] major qualifier.
[34:27] >> Sure.
[34:28] Uh medieval
[34:31] >> Medieval times?
[34:33] I think we lost him.
[34:35] >> His internet went back
[34:38] >> to 1988.
[34:39] >> growing food but in a very different way
[34:42] from how things work now.
[34:43] >> Yeah.
[34:44] Uh well, thank you so much for taking
[34:46] the time to come chat with us. It's
[34:47] always a pleasure to go all around the
[34:49] world topics.
[34:51] >> Uh have a fantastic week and a fantastic
[34:53] summer.
[34:54] >> all next time and work like I said, work
[34:56] harder.
[34:57] >> Yes.
[34:57] >> When you're next time you're in
[34:58] California
[34:59] >> Yeah, come to our studio.
[35:01] >> a cup of Joe. We will try to We will try
[35:04] to create an experience
[35:05] >> of coffee in history
[35:07] >> That would be
[35:08] >> do a 4-hour show.
[35:09] >> incredible
[35:10] >> Tyler Cowen
[35:11] >> me to do it.
[35:12] >> Let's see. We'll peer pressure him.
[35:13] >> very hard.
[35:13] >> think it'll work. Anyway, thank you so
[35:15] much Tyler.
[35:16] >> Cheers.
[35:16] >> We'll talk to you soon.
[35:17] >> Bye.
[35:17] >> Bye.
[35:18] >> Have a good rest of your day.
