Marginal Revolution

Zero-sum thinking, not DeepSeek's specifics or China-threat perception, best explains why Sputnik triggered American investment while DeepSeek triggered retrenchment. Sputnik brought more education funding, ARPA, R&D spending, loosened immigration, and falling tariffs; the DeepSeek response has run nearly opposite. Cowen weighs four rival explanations from readers. Dismissing DeepSeek as unimpressive misses that it symbolized China's broader rise as the world's largest exporter, manufacturer, electricity producer (a chart shows China's electricity generation dwarfing other nations), and military by headcount. Existential-threat framing fails too: a greater threat would likely provoke harder restriction, not liberalization. Rising median voter age (30 in 1950 to 39 now) doesn't fit either, since the anti-China Trump coalition is angry and disruptive, not risk-averse. Fiscal and political slack matters less than changed preferences. Instead, citing Chinoy, Nunn, Sequiera and Stantcheva's NBER research, Cowen argues growing zero-sum beliefs -- that one group's gain is another's loss -- now drive protectionism, anti-immigration sentiment, and hostility to university and science funding. Davidai and Tepper find zero-sum beliefs rise under perceived threat and assertive leadership and correlate with anger, distrust and populism; they're lower among immigrants and older Americans who've experienced growth. The result: a self-fulfilling "zero-sum trap," where zero-sum beliefs produce zero-sum policies that produce zero-sum outcomes.

Europe's weak negotiating position in the EU-USA trade deal reflects structural weaknesses in its economic and political model, not poor negotiating skill. Olivier Blanchard and FT commentators call the deal raw for Europe; Cowen treats that as evidence of Europe's fragility, while crediting the Trump administration with a "remarkably good job" by its own standards. Justin Wolfers objects that Trump is raising taxes on Americans (Cowen agrees), yet Europeans get modestly lower taxes as a result, even as EU leaders protest — and Europe's own non-tariff barriers remain significant. Conor Sen notes the tariffs have become a multi-trillion-dollar revenue source without visibly hurting corporate profits. Cowen expects the tariffs to persist, functioning as an early step toward resolving America's fiscal position, bringing higher inflation and lower real wages rather than fiscal collapse. He'd have preferred cutting subsidized health care instead, but concedes health care "polls well" — a point he says implicitly puts Matt Yglesias closer to pro-tariff politics than Yglesias would like. A future Democratic administration, he predicts, will spend the tariff revenue rather than repeal it, since Democrats have long been lukewarm on free trade despite their current anti-Trump rhetoric. The upshot: revise estimates of the political equilibrium.

A uniform tariff on all trading partners can produce smaller net losses than a selective tariff on just one country, even though the selective tariff looks closer to free trade. Cowen illustrates this with an example (adapted from Malaysia economist Apurva Sanghi) where the US can buy Hyundai Sonatas from Korea ($40k) or Toyota Camrys from Japan ($43k), perfect substitutes for 50 buyers. Under free trade, everyone buys the cheaper Sonata. A uniform 10% tariff raises both prices (Sonata to $44k, Camry to $47.3k) but buyers still pick the cheaper Korean car: consumers lose $200k while government gains $200k in revenue — a wash nationally, since it's just a transfer. A selective 10% tariff on Korea alone leaves the Camry untaxed at $43k, so buyers switch to the costlier Japanese car — trade diversion. That generates zero tariff revenue and a real $150k national loss, since production costs rise with no offsetting transfer. Cowen likens this to taxing GM but not Ford, letting a less-efficient producer win. Citing Brian Albrecht, he notes tariffs are generally a costly way to raise revenue, that selective tariffs are especially wasteful, and that the actual US tariff system — highly selective, with different rates by country and time — makes trade diversion near-inevitable while inviting rent-seeking lobbying for favorable treatment.

A new NBER paper by John Becko, Gene Grossman, and Elhanan Helpman argues that when great powers care about smaller states' political allegiance as well as economic welfare, optimal tariffs exceed the classic Mill-Bickerdike level. The authors model a unipolar world, where a hegemon uses preferential trade deals to attract allies, and a bipolar world, where two powers compete for influence and their tariffs can be strategic complements or substitutes. Calibrating with U.N. voting patterns, an estimated cost of buying U.N. votes, and military spending data, they find geopolitical motives significantly amplify protectionist pressure, and that the emergence of a second great power (China's rise) helps explain the retreat from globalization -- a point Tyler Cowen calls valuable and original.

Cowen otherwise objects that the model assumes a government maximizing only economic-plus-geopolitical welfare. He wants it extended to governments that also pursue corrupt side-bargains, maximize treasury revenue beyond the optimum for Leviathan-like reasons, or hold ideological, behaviorally ungrounded attachments to protectionism -- additions he says would change the results. He argues this gap is symptomatic: mainstream, non-libertarian economists keep producing research that implicitly defends protectionism while refusing public-choice critiques that would unsettle their other priors. Even incorporating corruption, Leviathan, and ideology, he adds, the analysis would still yield a strong case against Trump's tariffs.

Tyler Cowen argues the US and China have effectively swapped historical postures on attracting global talent, and America risks squandering its biggest advantage by abandoning openness. The US answered Sputnik with expanded science-education funding, new agencies like ARPA, higher federal R&D spending, and recruitment of foreign talent; its response to China's rise has run the opposite direction. China, meanwhile, just launched a K visa (announced August 14) for young science and technology professionals, offering more entries, longer validity, and streamlined applications requiring no domestic employer invitation — layered atop visa-free entry deals reaching 75 countries by July 16 (adding Azerbaijan), covering nearly all of Europe since a December 2023 start with France, Germany, Italy, the Netherlands, Spain, and Malaysia, plus five Latin American countries, Uzbekistan, and four Middle Eastern states, about two-thirds on one-year trials. The US, facing a shortage of high-IQ workers, instead treats immigrants as threats: visa-interview wait times now run months to years, international students are being deported and discouraged, and research funding is being cut, costing America ground in academic talent. Cowen concludes China still isn't good at attracting talent — the world's best still prefer America — but continued closure could cost the US its exceptionalism.

A few turns of history could plausibly have made China the world's most populous Christian nation, via the Taiping Rebellion (1850-1864), a civil war with 20-30 million dead—far exceeding the US Civil War's ~750,000—that Tyler Cowen calls modern history's most important event educated Westerners barely know. The revolt began with Hong Xiuquan's 1837 visions, following repeated failure at the Confucian civil service exams; in 1843, after a fourth failure, a Christian tract revealed to him that his visionary "elder brother" was Jesus and that he himself was God's second son. Hong's movement fused this millenarian Christianity with nationalist rage against the ruling Manchus, growing past a million fighters and seizing Nanjing as capital of the Taiping Heavenly Kingdom, which banned foot-binding, prostitution, and slavery, promoted gender equality, and distributed bibles. The kingdom appealed to Britain and America as Christian brothers, and one commander's letter (quoted from Platt's *Autumn in the Heavenly Kingdom*) called them "sons of the Heavenly Father." Britain ultimately backed the Qing instead—a choice Ito Hirobumi called Britain's worst China mistake—and internal purges doomed reformer Hong Rengan's modernization push. Sixty-three years later, Mao's revolution followed a strikingly similar pattern, but with Marxism-Leninism replacing Christianity, and, Cowen argues, far worse results.

Rare-earth "crises" recur roughly every decade, and each time supply proves more elastic than feared. In 2010 Paul Krugman warned the US had let China acquire a monopoly "exceeding the wildest dreams of Middle Eastern oil-fueled tyrants." Tyler Cowen argued in 2014 that this was overstated: Chinese producers evaded their own export bans, firms that had relied on cheap rare earths cut usage once prices rose, new suppliers entered, and innovation found substitutes.

The panic has returned, and Cowen cites Tim Worstall, a rare-earths dealer, who argues rare earths are "neither rare nor earths" and exist nearly everywhere. The real bottleneck is regulatory: the easiest ores, like monazite (a waste product of other industries), carry light radioactivity, and irrational rules block their use. Marco Rubio's proposed Thorium Act would unlock Florida's phosphate gypsum stacks, which hold substantial rare-earth quantities. The claim that only China has dysprosium/terbium ores (needed for high-temperature magnets) is also outdated — "ionic clay" ores once thought unique to South China and Burma can form wherever granite weathers in subtropical climates; Worstall cites a dozen Australian deposits plus active mining in Chile and Brazil. He stresses this isn't a case for decades of subsidy — capacity built now will be cheaper and more technologically current than anything propped up earlier.

Judged in strictly utilitarian terms, U.S. interventions in the Western Hemisphere that actually removed a country's leaders have mostly succeeded, the frame for assessing Venezuela today. Puerto Rico (1898) was a big success; the Mexican-American War removed Mexican leaders from the future American Southwest, benefiting even the Mexicans living there now; the coup against Allende in Chile succeeded, given Chile's current wealth and democratic stability; Grenada under Reagan was a modest improvement over Marxism; Panama, under the first Bush and earlier for the Canal, was a big success both times. Haiti (under Clinton, and 1915-1934) was a terrible failure despite an unclear counterfactual; Cuba (1906-1909) is simply unclear. The Dominican Republic, starting with Trujillo's removal, eventually succeeded but took a long time, while the earlier 1916-1924 occupation is not an obvious success. The American Revolution and Civil War both count as successes. Ecuador (1963) and Brazil (1964), where U.S. involvement was only partial, are unclear, as is Nicaragua (1909-1933). Two patterns emerge: Latin American interventions outperformed Middle East ones, likely from stronger regional ties to democracy and proximity; and successes took a long time and weren't quite the kind sought. A study by Absher, Grier, and Grier found poor results from CIA activism in Latin America, attributed to cases like Nicaragua and Cuba where leaders were never removed — merely funding conflict yields poor returns.

Davos-style commentary and Mark Carney both misread what's actually happening to free trade and to "middle power" autonomy. Trump's tariffs mark a genuine US retreat from free trade, which the author regrets — but the same trade war pushed Canada toward freer trade elsewhere: Canada has opened its market to Chinese electric vehicles, giving Canadians access to better, cheaper EVs than Americans can buy, even granting that spyware concerns might justify keeping such cars out of Ottawa specifically. Meanwhile the EU reached a pending free trade agreement with Mercosur, the South American bloc covering hundreds of millions of people, and is likely to strike one with India, the world's most populous nation and among its fastest-growing economies. The author asks whether any recent period has seen this much free-trade progress.

On Carney: his widely praised speech claimed middle powers like Canada and Europe can stand their ground against the US and China, but barely addressed AI, even though Canada is deeply dependent on American AI systems (often built by Canadian researchers working in the US) — soon to be the defining issue in the relationship. Carney can afford defiance only because America will defend Canada regardless, including against Russia; Europe, lacking that guarantee, stays deferential. The real pattern isn't rising middle-power independence but a choice between tethering to a major power or living in fear of abandonment.

The dissent's claim that IEEPA tariff authority is implied by the President's power to ban imports entirely gets delegation logic backwards, since the two instruments have opposite political-economy properties. The Supreme Court struck down Trump's IEEPA tariffs, holding "regulate... importation" excludes tariffs; the majority noted actual congressional tariff delegations specify "duty," cap rates, and impose time limits and procedures, which IEEPA lacks. The dissent countered that granting the "greater" power to prohibit trade implies the "lesser" power to tax it — echoing Trump's own claim that he can "destroy the country" but supposedly can't "charge a little fee."

Standard principal-agent theory refutes this: Congress wants emergency delegation to screen for genuine emergencies. A ban is a costly, self-limiting signal — it inflicts immediate harm, so invoking it credibly proves urgency. Tariffs are cheap for a president: they raise revenue offsetting political pain, their incidence is diffuse and easy to misattribute, and they're adjustable — useful for bargaining and targeted favors (Cowen cites Trump soliciting a "gold bar" from Switzerland for tariff relief). A fire chief's road-closure power illustrates the same asymmetry: it doesn't imply toll authority, since closures are self-limiting while tolls persist and build constituencies. Taxing power sits with Congress because revenue instruments are uniquely exploitable by an executive, being targetable.

Global funding for developing-world water and electricity infrastructure is declining, and Tyler Cowen argues the deeper problem is institutional: political incentives favor building over maintaining, and corruption erodes what gets built. Introducing a genuinely new institutional order, he notes, is harder and more perilous than any construction project. Connor Tabarrok's account of Ek Son Chan illustrates this: in 1993, Phnom Penh's water utility served only 20 percent of the city, ran 10 hours a day, and lost 72 percent of output to leaks and theft. As Director General, Chan fired corrupt managers, metered every connection, computerized billing to kill kickbacks, and raised bill collection from 48 to 99.9 percent, using worker bonuses and an internal discipline commission. He mapped pressure zones and ran nightly leak-detection patrols. Enforcement provoked resistance: when a general's bodyguards blocked a meter installation, Chan returned, was held at gunpoint, then brought 20 armed police to cut the general's water until he signed a public apology. By 2010, coverage exceeded 90 percent with 24-hour service; the utility turned a profit and listed on Cambodia's stock exchange in 2012. Chan won the 2006 Ramon Magsaysay Award. By separating the utility from weak local government, he showed that infrastructure depends on institutional quality and mechanism design — state capacity need not reside within the state itself.

The trade accounts are among the most misleading statistics ever collected, because they repackage millions of voluntary transactions between individuals and firms as a relationship between nations, as if "America" trades with "China." Even the word "deficit" biases perception, though a current account deficit is always matched by a financial account surplus.

Tyler Cowen highlights Don Boudreaux's (Cafe Hayek) reply to a reader who asked: since the US ran trade deficits for 50 years, offset by foreigners' investments, don't Americans eventually owe foreigners returns on that money? Boudreaux answers no. Only Americans who borrowed from foreigners (including government debt, ultimately owed by taxpayers) owe anything. Equity investments—like a foreigner buying a restaurant in Houston—carry no such obligation: if it goes bankrupt, the foreign owner alone absorbs the loss; if it profits, those dividends are new wealth the foreign owner created, not a payment extracted from Americans. That dividends flow abroad and get recorded as "payments" from America is a geographic accident mistaken for economic reality.

Cowen concludes the accounts commit a category error, treating a "where" as a "who." This isn't academic: politicians cite trade "losses" to justify hostile policy toward other countries. Bad accounting produces bad policy—we'd be better off without the trade accounts entirely.

Peter Thiel is right to relocate to Argentina, but for the wrong reasons: the libertarian revolution he sees in Milei doesn't exist as advertised. Since December 2023 Milei has cut inflation and produced a fiscal surplus, partly backed by Washington, but both supporters and critics have exaggerated his impact. Retenciones (export taxes on soybeans, wheat, corn that fund the welfare state and suppress domestic food prices) survive despite campaign promises, because Milei needs the revenue; union control of labor persists; reforms barely touch the Peronista-run northern provinces. The peso remains distrusted, the informal economy is 40-50% of employment, lines for Spanish/Italian citizenship stay long, and tycoons like Marcos Galperin still base themselves in Uruguay. Peronism endures as a social identity, not just a machine. Culturally, Argentine youth want Spain/Italy-style leisure and welfare, not a Protestant work ethic—yet PayPal, Palantir, and Facebook didn't come from Spain or Italy. Few young Argentines want to move to the US, and most Milei voters (many ex-Peronists, like Rust Belt ex-Obama Trump voters) were rejecting exhaustion with Peronism, not embracing libertarianism. Thiel's real case rests on demographics, human capital, lithium/rare earths, and isolation from great-power conflict—not civilizational renewal.

Mexico's drug violence reflects not a recent breakdown but the fact that it has never fully consolidated as a nation-state: one estimate puts drug gangs in control of as much as a third of its territory, continuing a pattern stretching back to independence. When Mexico broke from Spain in 1821, Central America joined briefly then split off, never sufficiently unified with the rest. Yucatán rebelled repeatedly afterward, nearly seceding; the U.S. seized roughly half of Mexico's territory in the 1840s war after Texas had already broken away; civil war followed in 1857; the French installed the Habsburg Maximilian as emperor by 1864, though he never controlled the country and was soon deposed and executed; the 1910 Revolution killed about 10 percent of the population. The calmer 20th century still left much of Mexico outside unitary rule, with remote regions seeing the state as oppressor rather than protector. When Colombian trafficking was suppressed in the 1990s, gangs filled the vacuum, becoming de facto rulers who supply protection and extract tribute — Mexican avocado exports, for instance, partly fund gangs. Mexico's central government spends under 23 percent of GDP, versus over 32 percent in Brazil despite comparable living standards — too small to defeat or preempt the gangs by force.

Iran's post-war behavior suggests critics of the US/Israeli strikes have not actually thought through what causal model of Iran their opposition depends on. Tyler Cowen recalls that before the war he had agreed with Matt Yglesias's case against attacking, and nothing in the war's course changed that view. Afterward, many commentators argued the resulting deal was a major win for Iran and a loss for Trump. But Iran is now attacking ships in the Strait of Hormuz again, openly discussing building nuclear weapons, and planning tolls or fees on Strait traffic — only the ship attacks actually surprise him.

Given this reckless behavior despite an ostensibly favorable outcome, Cowen asks what the correct model of Iran is: would they have behaved this way anyway absent the Trump/Netanyahu strikes, reflecting an underlying equilibrium (e.g., if drones have permanently shifted regional military balance, or if the military was always positioned to succeed the mullahs), or does Iran's current aggression reflect genuine path-dependence caused by the attack itself? War critics, he argues, implicitly assume the latter — that Iran would have stayed peaceful without the strikes — but rarely defend that assumption explicitly, and he questions what expertise justifies their confidence in it.

Tyler Cowen argues that America's edge over Europe will widen over the next 20 years, because most of the forces in play compound over time. America's worst problems — obesity and opioid addiction, which kills about 54,000 people a year — are improving: GLP-1 drugs are curbing obesity and opioid deaths have started declining. Coming biomedical advances will help Americans more than Europeans precisely because Americans start in worse health. America's scale advantage in tech should also grow: Nvidia alone, near a $4 trillion valuation, is worth more than the entire German stock market, while EU regulation actively discourages tech growth. Brain drain from Europe to US tech and AI is accelerating among the young; even though most won't have big hits at 22, more will by 40, so talent that has already left will keep paying larger dividends over the next 10–15 years. A cited NBER study found Europeans switch jobs far less often and undergo less restructuring — a static economy that falls further behind each year. Demographics compound the problem too: US fertility is about 1.63 versus the EU's 1.38, straining European pensions and leaving an older, less innovative population. Cowen notes he also covers immigration in the full column.

Market prices, not gut anxieties, are the most honest gauge of how bad things really are in Trump-era America, and by that gauge the alarm is overdone. High stock valuations trace partly to the One Big Beautiful Bill Act, which extended corporate tax cuts: the corporate rate fell from 35 percent pre-Trump to 21 percent in his first term, with 100 percent bonus depreciation and expanded interest deductions lowering effective burdens further. But federal debt now exceeds $37 trillion, so someone — likely today's taxpayers and their children — must eventually cover the gap, meaning elevated stock prices are partly borrowed from the future. The dollar has fallen roughly ten percent, signaling some loss of confidence, yet fears that Trump is destroying Federal Reserve independence and igniting inflation aren't showing up in the data: expected inflation measures mostly sit between 2 and 3 percent, several clustered near 2.5 percent, no different from late in Biden's term. On democracy collapsing — cited by Yale professors Jason Stanley, Timothy Snyder, and Marci Shore, who left for Canada — Kalshi's prediction market has Democrats favored at 53 percent for 2028, barely an edge, not a sign of fascism. If markets are wrong, critics should bet against them instead of merely predicting doom.

Central bank independence is less real and less important than commonly assumed, so market calm about Trump's anti-Fed moves — rhetoric, firing Lisa Cook, seeking a CEA seat on the board — makes sense, though the author opposes those moves. The Fed was never fully independent: in 2008 it coordinated crisis response with Treasury and other parties, and did the same during the 2020-2021 pandemic, both times ultimately deferring to a broader consensus (yielding too little inflation the first time, too much the second). Independence is most visible, and matters least, when stakes are low and roughly 2% inflation satisfies everyone; the real danger comes when government quality is poor, since encroaching on the Fed then just raises the level of stupidity, as is happening now — and looming fiscal dominance from recent budget decisions may erode independence further regardless. New Zealand shows a non-independent central bank can work fine: since the 1989 Reserve Bank Act it has had only operational independence, with inflation targets set jointly with government, yet inflation fell with political cover. Milton Friedman once favored having Congress set inflation directly — a bad idea, but proof independence isn't sacrosanct. Scatter plots correlating independence with lower inflation would likely lose that correlation if government quality were controlled for. Addendum: critics of Trump's moves, including Paul Krugman, had previously called for higher (3-4%) inflation targets themselves.

The 2025 Nobel goes to Joel Mokyr, historian of the industrial revolution, and growth theorists Philippe Aghion and Peter Howitt, known for the Schumpeterian growth model. Mokyr's Lever of Riches argues technological progress is historically exceptional, not typical: societies mostly slide toward stasis, forces opposing progress usually outweigh those favoring it, and today's prosperity owes more to accidental factors than commonly assumed—progress is fragile and can be halted by small environmental changes.

Aghion and Howitt's model, like Paul Romer's, makes growth endogenous rather than exogenous (as in Solow); but where Romer's growth is horizontal, from new product varieties, theirs comes from creative destruction—new firms and technologies displacing old ones. This links their framework to firm-level entry/exit data pioneered by Haltiwanger. A chart (Bartelsman et al.) shows the covariance of firm size and productivity: strongly positive in the US, under half that in the UK (1993-2001), and negative in Romania, where large state firms were least productive, until market reforms shifted labor to efficient firms.

Creative destruction requires some firms to fail and workers to be displaced. Business dynamism has declined in both Europe and the US, though the US retains a more flexible labor market and is more open to firm births and deaths. Worldwide, productivity dispersion between firms is rising, meaning good ideas are spreading more slowly than before.

US federal debt keeps growing with no repayment plan, yet Treasury interest rates remain puzzlingly low — a paradox Tyler Cowen and Alex Tabarrok examine on the Marginal Revolution Podcast episode "America's Debt: Crisis or Calm?" Tabarrok cites Hanno Lustig and coauthors' NBER paper "US Public Debt Valuation Puzzle": debt gets paid down when the economy is strong (GDP high, marginal utility of money low) and grows when times are bad — the opposite covariance from gold, which hedges a portfolio by rising in downturns. Because debt is positively correlated with good times, he argues, its interest rate should logically be much higher than it is.

Cowen reframes the puzzle around stocks rather than flows: US wealth runs roughly six to eight times GDP (a rough figure, hard to value and illiquid), and in a peaceful economy that keeps building durable, undestroyed capital, the wealth-to-GDP ratio tends to rise over time. At 100% debt-to-GDP, that implies a debt-to-wealth ratio of only about 20% — comparable to a mortgage-to-net-worth ratio nobody worries about. Tabarrok notes the US, unlike an individual, is effectively long-lived; Cowen extends the logic: even 200% debt-to-GDP implies just 40-50% debt-to-wealth — high but tolerable, similar to mortgage burdens he's personally carried.

Two underrated factors explain more of a country's or region's macroeconomic trajectory than most analysis credits: human capital (how much active, ambitious talent exists, and how high the averages and medians run) and matching market demand (whether production is geared to what markets, especially export markets, actually want). Cowen cites these as the main reason he rejects "the housing theory of everything," and as the biggest drivers of the UK's current economic trouble: brain drain has hit human capital as smart Britons leave for the US or Dubai, while deindustrialization, the fading promise of business consulting, and London's decline as a financial centre (damaged by 2008, Brexit, and better innovation elsewhere) have hit demand-matching. Neither factor is easily fixed by reform. This framework also explains why growth rates show little serial correlation decade to decade: human capital shapes levels more than growth rates, and demand-matching hinges on luck and shifting consumer preferences that supplying countries can't always follow (Japan's carmakers cannot easily pivot as buyers switch from Toyotas to Chinese EVs). Institutional theories, by contrast, over-predict serial correlation; institutions matter mainly by shaping #1 and #2 long-term. An addendum credits Fischer Black, underrated beyond his options-pricing work, as an influence.

The French aren't lazy so much as unambitious at a social level, and that gap in ambition — not leisure preference itself — is what's worth criticizing. Economist Olivier Blanchard opens the debate on X, arguing the French simply value leisure more and are rationally converting productivity gains into fewer hours and earlier retirement rather than income; anyone who wants to work 50-60 hours and retire late still can, though most jobs aren't exciting enough to make that appealing. John Cochrane and JFV also weigh in. Tyler Cowen agrees "lazy" is the wrong frame: West Europeans produce good work per hour but choose fewer hours than Americans or East Asians, driven partly by taxes (themselves endogenous to social mores) and partly by genuinely better leisure options than much of the US — though pre-1970s Europeans worked longer hours, so taxes aren't the whole story. Cowen's real complaint is ambition: the world may be in a period of increasing returns (China and the US pulling ahead), and over the long run, stacking General Purpose Technologies always produces such periods. France contributes in science and math and invented an early internet, but could do more, and Cowen sees no offsetting externality from French leisure — unless it induces Americans to work harder to afford vacations there.

Minimum wage laws for gig workers fail to raise take-home pay because platform labor supply is highly elastic, so mandated per-task pay increases get absorbed by more idle time rather than higher earnings. Cowen's 2017 "Uber Tipping Equilibrium" argued that since almost anyone with a decent car can drive, driver-hour supply is essentially perfectly elastic at the market wage; when tips raise net per-ride pay, more drivers enter, each finds riders less often, and paid driving time falls until net hourly pay returns to baseline. A NBER paper by Hall, Horton and Knoepfle confirmed this empirically.

The same logic applies to Seattle's 2024 "PayUp" law, which set a minimum pay rate for gig workers like DoorDash couriers based on time and distance per task—a floor for one worker type within a larger market, not a universal minimum wage. A new NBER paper by An, Garin and Kovak found the predicted result: pay per task rose, but highly attached incumbent drivers completed fewer tasks (without more exiting delivery work), completely offsetting the per-task gain and leaving monthly earnings unchanged; drivers instead saw more unpaid idle time and longer distances between tasks. Their model attributes this to free entry pushing down the task-finding rate until expected earnings return to pre-reform levels—an instance of the general "Happy Meal Fallacy."

Europe's strict dismissal law doesn't just dampen hiring at the margin — in general equilibrium it reshapes which firms get built at all, steering capital away from risky ventures and explaining why Europe hasn't produced a Tesla. Cowen and Tabarrok's textbook analogy: requiring marriage for every date makes dating harder, just as requiring lifetime commitment for every job makes hiring harder. Brian Albrecht extends this: even a firm eager to hire, like Siemens facing a demand surge, hesitates because firing later is costly — and workers are locked in at firms like BMW that won't let them go either, choking off supply. Pieter Garicano details Germany's Kündigungsschutzgesetz: layoffs of 10+ require a "social selection" ranking by age, tenure, family obligations, and disability; disabled workers need Integration Office approval; caregivers are protected from dismissal for two years; works councils (required above 5 employees in Germany, 50 in the Netherlands) must approve layoffs; collective dismissals above 30/month trigger negotiated "social plans." Opel's Bochum closure cost €552 million for 3,300 workers, including payments up to €250,000. Result: firms favor stable, regulated business over radical experimentation — EV leaders are new entrants (Tesla, BYD) or gutted incumbents (MG), not established European automakers.

Eliminating Social Security's wage cap while capping benefits — a reform pushed from the center-right on grounds that the system is insolvent and needs "tough choices" — moves Social Security in exactly the wrong direction. One of the program's best features is that it has never been purely redistributive: benefits track earnings history, so people who paid in more get larger checks. Severing that link turns Social Security from forced saving into welfare financed by an uncapped tax on earnings.

This also worsens marginal incentives. A tax that buys a claim on future benefits isn't a pure tax — putting 10 percent of salary into a 401(k) cuts current consumption but returns an asset, making it closer to a purchase than a tax, and a reason to work more rather than less. Because Social Security embeds that same logic, the payroll tax is currently less distortionary than an equivalent flat tax. Break the contribution-benefit link and the tax grows more distortionary even if total taxes and total benefits are unchanged.

The better reform runs the opposite way: strengthen the tie between contributions and benefits, making Social Security more like an individual account. This also defuses the politics — a claim-based system, unlike a welfare program, isn't zero-sum. Addendum: James Buchanan made this argument first; John Cochrane endorses the economics.

Two independently estimated NBER papers give a consistent account of California's $20 fast food minimum wage: it raised wages, cut jobs, and pushed up prices, with price pass-through — not automation — driving the job losses. Clemens, Edwards, and Meer find wages in the sector rose about 8 percent relative to the rest of the country, while employment fell 2.3–3.9 percent (median ~3.2 percent), roughly 18,000 jobs. Clemens, Edwards, Meer, and Nguyen find food-away-from-home prices in California's four CPI-reporting metro areas rose 3.3–3.6 percent versus 17 control metros, with falsification tests showing zero differential movement in food-at-home or core inflation — the effect is specific to restaurants. Applying Andreyeva et al.'s -0.8 demand elasticity to the price increases implies quantity declines of 3.9–4.1 percent in limited-service and 1.7–1.8 percent in full-service, closely matching the employment estimates. Welfare-wise, most workers gained an 8 percent raise while a minority lost jobs, funded by a transfer from California's 39 million consumers — likely regressive, since lower-income households spend more of their budget on fast food. Small minimum-wage increases may be absorbed, but the effect is non-linear: large increases bite.

Attacking Jeff Bezos for calling the U.S. tax system progressive, liberal economists undercut eighty years of their own doctrine on payroll taxes. Bezos tweeted that the top 1% pay 40% of taxes, the bottom 50% pay 3%, and proposed zeroing out the bottom half's taxes to make the system still more progressive. Gabriel Zucman and Justin Wolfers shot back that regressive payroll and consumption taxes make the system less progressive than Bezos claims—an "own goal," Cowen argues, since Bezos was proposing a cut, not defending the status quo.

Worse, treating payroll taxes as regressive taxes contradicts the liberal framing running from FDR ("no damn politician can ever scrap my social security program") through Arthur Altmeyer, Wilbur Cohen, Robert Ball, and Nancy Altman/Eric Kingson's book Social Security Works!, which insists benefits are "earned compensation," not welfare.

Cowen proposes a Buchanan/Larry Summers test: a mandated payment is a tax only if it exceeds the payer's marginal benefit. Medicare (HI) buys zero marginal benefit, so it's a progressive tax (2.9%, rising to 3.8% above $200K/$250K). Social Security's bend-point formula gives low earners better returns, so OASDI is a contribution at the bottom, a tax at the top. Abandoning the "earned right" framing, Cowen warns, hands privatizers at AEI and Cato their argument.

The Coase Conjecture — that a durable-goods monopolist's inability to commit forces prices to unravel to marginal cost "in the twinkling of an eye" — fails decisively against real-world data. Coase's original six-page paper spawned formal proofs by Gul, Sonnenschein and Wilson and by Fudenberg, Levine and Tirole showing the logic holds under general conditions, an odd result since it implies patents and copyrights are nearly worthless. In a new paper with Tim Groseclose, Tyler Cowen tests the conjecture using e-book prices — a clean setting given low marginal cost, limited resale, and fast repricing — using public-domain e-book prices as a marginal-cost proxy. Prices don't fall rapidly to MC, the market doesn't clear in period one, sales continue over many periods, and prices don't even decline monotonically. They rule out rising marginal cost and finite-buyer "Pac-Man" equilibria as explanations. Two candidates remain: sellers committing not to cut prices (Cowen's preference, arguing firms routinely commit to workers and contractors, so why not price) or Board and Pycia's outside-options model, which Groseclose favors — where consumers who leave the market never return, paradoxically giving firms a time-consistent incentive to hold high prices. Either way, the theory fails even in the setting most favorable to it.

"Forced heirship" laws across much of Europe require owners to bequeath 50-75% of their estate to their children, sharply limiting the ability to direct wealth to charity or foundations -- unlike the US, where owners can give as they choose (aside from spousal claims); Louisiana is the only US state with forced heirship, and even it gutted the rule in 1995. Citing a post by John Arnold, Tyler Cowen notes that were Arnold European, he would owe 75% of his wealth to his three children rather than to the philanthropy he and his spouse now fund. Arnold argues the laws reflect Europe's aristocratic heritage versus America's self-made ideal, applying even when heirs are estranged, addicted, or irresponsible. Crucially, French law counts lifetime gifts against the required share: with three children, 75% of assets at death plus prior gifts must go to them, so giving away more than 25% while alive exposes charities to lawsuits from heirs demanding the funds back -- stunting Europe's nonprofit sector. Spain, France, and Italy thus entrench family dynasties while the US uses estate taxes to limit them, yielding weaker philanthropy, civil society, and economic dynamism in Europe. Cowen adds that Piketty's Capital praises equal-division inheritance as an egalitarian revolutionary legacy without noting it blocks entrepreneurs from giving fortunes to charity -- which Cowen calls a case for laissez-faire.

Britain's economic and institutional decline, chronicled in Idrees Kahloon's Atlantic piece, is a warning for the United States, which increasingly resembles a declining Britain propped up by one or two dynamic sectors like AI. Kahloon's anecdotes: the NHS now spends more settling maternity-malpractice claims than on maternity care itself; a 2023 survey found one in 10 Britons doing DIY dental work, including pulling their own teeth. Incomes are low — junior doctors, striking for the 15th time in three years, start at £38,800; median civil-service salary is £35,680 — while PM Keir Starmer was mocked for a £200-a-night Valencia hotel during the Iran conflict. Infrastructure has become nearly impossible to build: HS2 rail, proposed in 2009, has tripled in cost to over £100 billion (a bat-protection structure alone needed 8,000 permits and cost £216 million), with the truncated line not finished until 2040. Hinkley Point C, the world's costliest unbuilt nuclear plant, has a 31,401-page environmental review and a £700 million "fish disco" to scare fish from its intakes. Cowen notes Britain and the US both respond with retrograde scapegoating — Brexit, tariffs, immigration attacks — rather than fixing the real causes of stagnation; he calls the abundance movement one of the few hopeful signs.

Samuel Scheffler's *The Mysteries of Time and the Goods of Attachment* argues against Derek Parfit's "time neutrality," the view that pleasure or pain isn't more or less valuable for arriving earlier or later. Scheffler's counterexamples: someone who will write six good books cares how many lie behind versus ahead, and a mourner wants grief to run a set length, front-loaded rather than open-ended. Cowen weighs three defenses of neutrality: a ceteris paribus clause (strict enough, it makes neutrality tautological); confining neutrality to impersonal social-welfare judgments rather than personal decisions (his own stance in *Stubborn Attachments*); and holding individuals should merely be more neutral "at the margin." He presses further — what discount rate lets Scheffler call some counterexamples compelling and not others? A rate of zero smuggles time neutrality back in. Cowen's own view: a multiplicity of not-fully-commensurable perspectives, including time neutrality (especially for social choice), is needed for a meaningful life — not where Scheffler lands. He also flags Scheffler's case for "finding meaning in the whole," an implied obligation to create meaning and attachment for others, and "archived lives," which Cowen thinks leans too heavily on one neurologically specific view of memory.

Sven Beckert's 1,103-page *Capitalism: A Global History* is erudite and well-written, but suffers two major flaws. First, "capitalism" is a weak organizing frame for centuries of material: it's unclear what actually unifies the discussions beyond a "replacement-level" left-wing set of cliches, and coverage is uneven — the final chapter's treatment of Pinochet's Chile and neoliberalism reads like ordinary NYT-level journalism rather than original scholarship.

Second, the book never pins down what capitalism actually is. It describes mid-17th-century Barbados (p. 180) as near-ideal, "almost perfectly Smithian" capitalism, yet King Charles I had granted all the land to the Earl of Carlisle to distribute (statism), the economy ran on slavery (which Smith condemned as both immoral and inefficient), and the 1651 Navigation Act restricted trade with the Dutch — mercantilism built on land theft and bondage, not capitalism.

The epilogue similarly holds up Cambodia as capitalism's contemporary embodiment, despite its ruling Khmer Rouge history, current dictatorship, ranking 98th ("mostly unfree") on the Heritage Foundation's economic freedom index, and dependence on Chinese domination — Poland, Ireland, Singapore, or the Dominican Republic would fit far better. The deeper issue, the reviewer concludes, is that the author doesn't know what he doesn't know — symptomatic of how insular elite higher education has become.

Critics and defenders of liberalism alike tend to get their numbers and history wrong, and American liberalism itself has never been pure but has always leaned on illiberal means. Reviewing Philip Pilkington's critique for the TLS, Cowen finds the empirical claims collapse: Pilkington blames liberalism for "atomized" families unable to "pass on their genes," ignoring that fertility is falling even faster in illiberal Russia and China; his line that "deindustrialization eviscerated American industry" ignores that US manufacturing output sits near its pre-financial-crisis peak, with service jobs now often paying more than manufacturing ones. Pilkington also borrows trade-imbalance theories from non-economists Oren Cass and Michael Pettis, holding that Chinese and German surpluses against US deficits demand major restructuring—a view mainstream economists reject as no less stable than any other macroeconomic arrangement, without explaining why the consensus is wrong.

On Robert Kagan, Cowen argues American history shows liberalism relying on illiberalism repeatedly, not accidentally: Lincoln abused civil rights including habeas corpus, and Union campaigns killed many civilians beyond military necessity; FDR tried to pack the Supreme Court, expanded executive power in ways prefiguring Trump, and firebombed Tokyo while fighting for liberal victory. Today's illiberal drift, Cowen concludes, looks like the normal ebb and flow of American politics rather than a decisive break from liberalism.

Britain's 1833 Slavery Abolition Act, which freed over 800,000 enslaved people across the Empire, was made possible only by paying slaveowners a £20 million bribe/compensation—about 5% of GDP, equivalent to roughly $170 billion in today's economy—debt not fully repaid until 2015, prompting outrage when revealed. Cowen argues compensation should ideally have gone to the slaves, but given that slavery had already happened, paying owners was the only peaceful path to abolition, as Wilberforce affirmed on his deathbed. A companion enforcement effort, the 1807 Slave Trade Act, was backed by Royal Navy patrols—illustrated by an image of HMS Black Joke firing on the Spanish slaver El Almirante—that peaked at over 14% of the fleet in the late 1840s-1850s, capturing 1,600 ships and freeing 150,000 people, yet proved only modestly effective on its own. The real turning point was Brazil's 1850 abolition of the slave trade—it had absorbed nearly 80% of trans-Atlantic slave voyages by mid-century—spurred partly by the 1845 Aberdeen Act, which let Britain seize Brazilian slave ships. A new paper by Yi Jie Gwee and Hui Ren Tan finds that votes to keep funding the costly, economically unrewarding patrols—maintained even through the Napoleonic, Opium, and Crimean Wars—were driven by ideology, not interest, a conclusion Cowen endorses as evidence of British moral commitment.

A new edited volume, Derek Parfit: His Life and Thought (ed. Jeff McMahan), gathers accounts of the philosopher from those who knew him, anchored by a memoir from his widow, philosopher Janet Radcliffe Richards. She recounts his severe insomnia while writing his book: he was sometimes awake thirty-six hours straight and confided that if a gun had been at hand he might have shot himself — not wanting to die, but desperate to lose consciousness. A sleep regime and calming drugs eventually solved it, after which he evangelized the fix to anyone with similar trouble, including her. Other details: he exercised only on a stationary bike, since it let him read simultaneously, and he loved Mitsuko Uchida's Mozart. Radcliffe Richards also describes what the relationship cost her own life: being known chiefly as "his partner," a social circle that contracted rather than widened, recreation limited to what he wanted, and no career help, since he was neither her teacher nor her referee. Larry Temkin recalls that Parfit couldn't calculate a 15% tip; Parfit's brother also contributes an essay. Cowen calls the contributors' collective devotion to their subject unmatched by any volume he's seen, and says it's worth its triple-digit price.

Opposition to scientific consensus is not simply uninformed—it is correlated with knowledge overconfidence, making the loudest dissenters the most confidently wrong: the people who know the least but believe they know the most.

Tyler Cowen cites a paper by Light, Fernbach, Geana, and Sloman that sidesteps a circularity problem (dissenters might sometimes be right) by first scoring respondents on uncontroversial true/false science questions, e.g. "antibiotics kill viruses as well as bacteria: false." They correlate these objective-knowledge scores with opposition to consensus on vaccination, nuclear power, and homeopathy. A chart plotting the two shows respondents most opposed to consensus score lowest on objective knowledge yet report the highest subjective confidence—the gap between what they know and what they think they know is widest among the most anti-consensus group. A follow-up test confirms the sincerity of this overconfidence: the confidently wrong bet more heavily on the objective questions, and lose—"a bet is a tax on bullshit."

The implication: the "knowledge deficit" fix (just giving people more facts) fails when the least-informed are also the most certain. The authors suggest leaning on social norms and trusted community figures instead. Cowen adds that classrooms, where information flow is clearer, see less confidently-wrong pushback than Twitter or blogs.

LDS theology's core commitments generate libertarian politics almost by structural necessity, argues Connor Hansen, summarized here by Tyler Cowen. LDS metaphysics holds that God operates within eternal, discoverable laws, echoing the Enlightenment view of a lawful, rationally knowable universe. Its epistemology is pro-reason: humans possess eternal "intelligence"—personhood and rational capacity existing independently of creation and surviving death—paired with genuine agency, yielding an optimistic anthropology where reason is a divine gift rather than corrupted. Agency is the most explicitly libertarian strand: in LDS scripture, God's plan for mortal life required real freedom to choose and err; Satan proposed instead eliminating agency and guaranteeing salvation through compulsion, and was cast out for it. Coercion and forced conformity are thus not merely bad policy but theologically Satanic. The doctrine of eternal progression—that humans can literally become as God over infinite time—makes systems that constrain human aspiration feel spiritually suspect. Historically, LDS expulsions from Missouri and Illinois, Joseph Smith's violent death, and the migration to the Great Basin reinforced suspicion of centralized power, while the modern LDS welfare system models voluntary, covenant-based mutual aid as an alternative to compulsory state provision. Cowen points readers to Hansen's full essay and the book "Latter-day Liberty," introduced by Mark Skousen.

As capitalism advances, the gap between cardinal utility (what makes you happy) and preference satisfaction (what you want) widens, and it's unclear whether that's good or bad. Happiness and preference diverge for reasons beyond addiction: people want to be a certain kind of person, want to help the world without expecting it to make them happier, or want children despite mixed evidence children increase happiness. In a primitive economy where rice is the only good, the two coincide perfectly, since there's nothing else to want. As product choice and income grow and basic needs are met, people gain more room to deviate from maximizing cardinal utility even as more of their preferences get satisfied. This isn't simply a loss, since getting what you want could itself be central to the good life, though some preferences are harmful addictions or self-stressing (e.g., compulsively checking texts, which need not make you happier but can be a legitimate preference, like a mother checking her daughter got home safe). Books condemning capitalism for this gap will proliferate, mostly misframed but "with a point," and collectively they erode the view of humans as agentic. Cowen instead sees markets as an engine for making people weirder and pushing them into different worlds.

Ludwig Lachmann — a Jewish refugee from Germany who became a South African economist at the University of the Witwatersrand — shaped Tyler Cowen's early economic thinking when Cowen, at seventeen in 1979, talked his way into Lachmann's NYU graduate seminar. Tipped off (possibly by Richard Ebeling) that Lachmann was worth hearing, Cowen introduced himself and was let in; Lachmann, then 73, had an Old World manner and a distinctive South African accent. Though visiting NYU's Austrian program under Israel Kirzner, Lachmann's seminar ranged beyond Austrian economics: Sraffa and Garegnani, the German historical school (via his own teacher Werner Sombart), and a reading of Keynes that pulled Cowen away from hydraulic IS-LM models toward cost-based pricing and G.L.S. Shackle's "kaleidic" economics. Lachmann held that markets had no overall tendency toward general equilibrium, though most individual markets did tend to clear, and he attacked W.H. Hutt's reading of Say's Law — prompting Cowen to read Hutt directly and absorb Hutt's critique of apartheid. Neither Lachmann nor Kirzner (also South African-rooted) ever discussed South Africa itself. Cowen still draws on Lachmann's Capital and Its Structure when thinking about AI economics, and remains grateful Lachmann let "a kid" sit in.

Economics is no longer defined by its subject matter but by who does the work and how rigorously, Tyler Cowen argues in an excerpt from his book "The Marginal Revolution: Rise and Decline, and the Pending AI Revolution." The field has stopped policing its boundaries with neighboring disciplines: he cites a paper on confidence gaps between men and women (Exley and Nielsen, 2024), published in the American Economic Review, that could pass as sociology, psychology, or gender studies; a paper estimating church attendance from cell-phone data (Pope, 2024); and Steven Levitt's well-known studies of baby-name choices and Sumo-wrestling corruption. This isn't deliberate "economic imperialism" — the operative standard is simply whether a paper is good enough to publish. What now distinguishes economics, per Cowen, is higher standards, harder work, better math, and higher IQs, not marginalist reasoning itself, though most current work remains consistent with it.

Training is shifting accordingly: advisors increasingly steer prospective PhD applicants toward majoring in math or computer science rather than economics, on the view that economics can be picked up in graduate school while quantitative grounding cannot. Top programs also increasingly expect a "predoc" apprenticeship with an established researcher before admission.

Financial economics has drifted away from marginalism and toward pure machine-learning calculation, to the point where its most successful current research barely qualifies as economic theory. The turning point: a 1992 paper by Eugene Fama and Kenneth French found that Beta, the centerpiece of CAPM, had no real power to predict expected stock returns — a devastating result since Fama himself helped build CAPM-style reasoning. In the most rigorous, highest-paying subfield of economics, marginalist theory has been retreating for over 30 years with no sign of reversal. Practice is already dominated by non-economist quants, and research is following: a 2024 Journal of Financial Economics paper by Scott Murray, Yusen Xia, and Houping Xiao ("Charting by Machines") shows theory-free machine-learning forecasts built from historical price data strongly and stably predict future returns — even among the 500 largest stocks, distinct from momentum and reversal effects — challenging the efficient market hypothesis. More strikingly, a working paper by Antoine Didisheim, Shikun Ke, Bryan Kelly, and Semyon Malamud extends Arbitrage Pricing Theory (traditionally five or six intuitive factors like inflation or interest rates) using machine learning to find 360,000 factors, cutting pricing errors by 54.8 percent versus the classic Fama-French six-factor model. No economic intuition can attach to 360,000 factors.

Most critiques of "capitalism" are really critiques of modernity — the organization of society around technology, formal institutions, and rational, impersonal rules — and conflating the two dooms reform proposals that misdiagnose the cause. Economist Jesús Fernández-Villaverde, in a Twitter/X series relayed by Tyler Cowen, argues that pathologies blamed on capitalism (pollution, fossil-fuel reliance, inequality, alienation, consumerism) appeared just as strongly under socialism: a Leningrad factory in 1970, a Jiangsu collective farm in 1978, or Leipzig couples in 1982 pursuing conspicuous consumption as eagerly as Chicagoans in 2026. De Beers didn't invent spending on engagement rings; it channeled a pre-existing evolutionary disposition. Calling the USSR or China "not truly socialist" is dismissed as a no-true-Scotsman dodge.

A second post makes the case concrete: imagine a socialist commonwealth trying to deliver state-of-the-art medicine — mRNA vaccines, MRIs, radiotherapy machines. Scaling that requires large bureaucratic organizations regardless of who owns property; scarcity forces denials regardless of system; and "more participation" or "more democracy" doesn't fix bureaucracy, as Yugoslavia showed and Arrow's impossibility theorem predicts. Conclusion: Marx got modernity wrong, Weber got it right — the bureaucratic "iron cage" is not capitalism's product but modernity's universal mechanism for coordinating large-scale tasks under rational, impersonal rules.

Jacobin, the DSA-aligned magazine, published an interview with sociologist Vivek Chibber (by Melissa Naschek) that lays out essentially the Mises-Hayek-Kornai critique of central planning: information problems and incentive problems doom planned economies, and supercomputers or AI cannot fix this, because if managers have incentives to lie, the input data feeding any algorithm is junk. Chibber refuses to blame Stalin, poverty, or "Russian backwardness" for planning's failures, treating the problems as structural rather than historical accidents. He even grants a point startling for Jacobin: under capitalism, managers are driven by profit to sell the best-quality good at the lowest possible price. His conclusion: the burden of proof is on the Left to show that replacing markets with planning can work, and dismissing decades of planning failures as not matching "my vision of socialism" guarantees repeating the same mistakes. Tyler Cowen adds his own critique -- Stalin was a consequence of central planning, not merely an impediment to it, and calling the outcome "the same dilemmas" euphemizes dictatorship, famine, and mass murder -- but agrees with Chibber's closing demand that the Left be "remorseless and merciless when it comes to facts," adding: replace "the Left" with "we."

A career runs about 80,000 hours -- 40 hours a week, 50 weeks a year, 40 years -- yet career decisions get far less rigor than choosing a mortgage, often boiling down to "follow your passions," which fails because too few jobs exist in the music, arts, and sports fields people name as passions. Benjamin Todd's newly updated book, 80,000 Hours, instead runs the numbers seriously. Framed around Effective Altruism, it argues pandemic risk is badly underfunded: an engineered pathogen killing over 100 million people this century is plausibly as likely as large-scale nuclear war or 6-degree climate change, and could kill over 90% of humanity, yet total spending on preventing worst-case pandemics is under $10 billion (just $1 billion philanthropic) versus climate change's $6-10 billion philanthropic and $1.6 trillion total. Despite the EA framing, most chapters serve purely selfish readers too, covering skills, networking, and job-hunting. Todd splits careers into explore, build, and deploy phases -- most people under-explore, and job-hopping while young is fine if it builds skills rather than pads a CV. He favors cheap, escalating tests over armchair pro-con lists: talk to LLMs and one person, then several more people and books, then a short project, and only then a multi-month commitment. Cowen calls it the Random Walk Down Wall Street of career advice.

Martha Nussbaum's *The Republic of Love: Opera & Political Freedom* is a strong book covering Mozart, Verdi's *Don Carlo*, Beethoven's *Fidelio*, and Adams's *Nixon in China*; her "political freedom" is looser than classical liberalism, though that difference is immaterial for some 80% of the book. Tyler Cowen, reviewing after Nussbaum declined a podcast, argues she subordinates the operas to her political agenda.

She brands *Don Giovanni* a "problem opera," denying women could desire the Don and blaming their choices on economic desperation — a rigid reading Cowen says Kierkegaard understood better. She's also hostile to romanticism, dismissing *Tristan* as "tedious" and "adolescent," with "no reciprocity," which Cowen disputes while granting most Wagner runs long. He suggests pairing her with Wayne Koestenbaum's *The Queen's Throat*, which reads opera's excess as politically charged for letting hidden — especially queer — identity surface, complicating any single political reading.

Cowen finds Beethoven less liberal than Nussbaum allows, given his "culmination" and communal-integration themes (Nazis later staged *Fidelio*), and faults the book's thin treatment of Rossini — opera's essence, yet a monarchist — noting Verdi too was a conservative monarchist, recasting *Don Carlo*. He ties this to Carl Schmitt on Romanticism's drift into reactionary monarchism, and to Mazzini and Auber's *La Muette de Portici*, which sparked Belgium's 1830 revolt. Verdict: thought-provoking and deeply knowledgeable, but classificatory, unwilling to follow where the music itself leads.

Marxian concepts—class, ideology, alienation, commodification—function as the default economics of literary criticism even in non-Marxist English departments, despite Marxism's failure as economics; a Hayekian/Austrian alternative is equally available and, Cowen argues, more accurate. The model is Paul Cantor's essay reading Thomas Mann's "Disorder and Early Sorrow," set during Weimar hyperinflation, through Mises and Hayek rather than Marx. Hyperinflation shortens time horizons: saving becomes foolish, so people flee into real goods, rewarding speed and risk-taking over prudence and thrift. This inverts generational authority—the elderly, fixed-income and inflexible, lose status to reckless youth, dramatized when Mann's children call their father by his first name and Professor Cornelius literally crouches to their height. Because money anchors a society's sense of value, monetary collapse shakes every other conviction, linking Weimar's nihilism to Nazism's rise: economic disorder becomes moral, psychological, then ontological disorder. Where a Marxist reading blames capitalism's contradictions, Cantor's reading blames state currency debasement—only one gets the economics right. Cowen proposes extending this method using spontaneous order, the fatal conceit, and the knowledge problem: War and Peace against the great-man theory, Brazil and The Lives of Others as the fatal conceit collapsing into absurdity, The Wire as a Hayekian epic exposing the illusion of social justice.

Rising child care costs are usually blamed on regulation, but comparing them to pet care costs shows the real driver is the Baumol effect. Prompted by paying $82/day to board his dog in northern Virginia (comparable to a Motel 6), Cowen compares BLS price series for pet services (boarding, walking, grooming) and child day care/preschool. The Institute for Family Studies links child care prices to regulations like staff ratios and training hours, but Cowen argues this misses the steady long-run rise, pointing instead to Baumol, as in his book with Helland, *Why Are the Prices So Damn High?* Two charts show pet and child care price trends are nearly identical over 26 years; adjusted for CPI, child care's real price rose 36% and pet care's rose 28% — since pet care is far less regulated, regulation can't be the main cause. This parallels Einav, Finkelstein and Gupta's finding that veterinary and human health spending rise at similar rates despite very different regulatory regimes. The mechanism: productivity stagnates in labor-intensive care work while it grows elsewhere, so wages (and prices) must rise to retain workers. This isn't bad for society overall, though it hurts heavy consumers of care and low-wage-growth workers; fixes include redistribution, low-skill immigration, and automation, though care resists mechanization.

Obamacare is probably not financially or politically stable, regardless of whether Paul Krugman's normative defense of exchange subsidies is right. Tyler Cowen traces the arc: before passage, advocates insisted all three "legs of the stool," especially the individual mandate, were needed to stop adverse selection and premium spirals—a view most economists accepted. The ACA passed by razor-thin margins with the mandate intact, which proved deeply unpopular, loading costs disproportionately onto exchange buyers and fueling Republican gains. Trump then effectively "saved" Obamacare by gutting mandate enforcement, letting people opt out penalty-free, which dissipated much of the backlash. Contrary to economists' predictions, the exchanges didn't collapse—enough people kept enrolling, plausibly due to positive selection. Under Biden, Democrats sustained premium support, though without much enthusiasm, given its cost. Cowen asks whether this allocation is sustainable long-term, or whether economists like Krugman and David Cutler were right all along, just with a lag—exchanges can't work without a mandate, and subsidies will keep growing costlier. He credits Megan McArdle's early warning that a policy passed by such a thin margin couldn't easily endure. In an addendum, Cowen notes that repealing Trump's tax cuts and taxing the rich won't come close to funding this; he suggests a VAT if that's the real answer, criticizing evasive posturing on financing.

US clinical trials cost so much primarily because US healthcare itself is expensive: sponsors must fund high-end care for every participant, and since American baseline care is already excellent, patients have little incentive to enroll — unlike in cheaper-healthcare countries, where the "level up" in care makes recruitment easier. Eli Lilly CEO Dave Ricks, on the Cheeky Pint podcast with John and Patrick Collison, quantified it: median enrollee cost is now $40,000, nearly two-thirds of the $60,000 median US wage. Roughly 20-30% of that funds superior patient care; the rest includes an institutional overhead premium, which Ricks likens to the "60% Harvard markup" debated in the NIH-funding cuts. Only 4% of US cancer patients enter trials, versus over 25% in Spain and Australia, partly because the US runs a decentralized IRB system requiring separate ethics approval at every site, while Australia has one national clearance. Prestige centers like MD Anderson and the Brigham draw the most trials but are also the most congested and slowest. Lilly's clinical-development timeline now runs about seven years, down from an industry average of ten, but half of that is spent waiting for enrollment; the company is testing database mining to proactively contact eligible patients. Cowen also flags an IFP essay on clinical-trial abundance and other podcast material on Eli Lilly's Zepbound-driven pharmacy business.

Trump's demand that U.S. drug prices match the lowest price paid in other developed countries won't cut American prices and risks making everyone worse off. Pharma's cost structure—a billion-dollar first pill, a 50-cent second one—makes price discrimination natural: charging poorer countries less still yields profit above marginal cost, funding more R&D, which benefits the U.S. too. But Americans see those lower foreign prices and feel cheated.

The policy strengthens pharma's hand abroad, since firms can now warn discounts will be re-exported into the U.S. Two problems follow. First, it won't lower current U.S. prices—firms already profit-maximize domestically, so a hike in France doesn't translate into U.S. savings; the real effect is higher pharma profits, a hard political sell. Second, forcing a single global price could backfire: since the U.S. generates most profits, firms may pick the high price, as Pfizer's CEO Albert Bourla suggested, telling France to pay more or lose access to new drugs. If countries like Switzerland—whose minister insists Swiss premiums won't subsidize U.S. cuts—refuse to pay up, they'll ration and delay drugs instead, as Europe already does. Likely outcome: falling pharma profits, slower R&D, unchanged U.S. prices, fewer new drugs abroad—lose-lose-lose.

Coercive pronatalist policy can produce a dramatic short-term spike in birth rates without durably raising how many children women actually have, evidence from Romania's 1966 Decree 770 suggests. Geruso and Spears's JEP survey on whether fertility can rebound examines the decree, which banned abortion and effectively blocked modern contraception. Figure 7, extending an approach from Sobotka, Matysiak, and Brzozowska (2019), adds Bulgaria as a control, since the two communist neighbors had tracked similar fertility trends before 1966. Panel A shows Romania's period total fertility rate nearly doubled the year after the ban, while Bulgaria's held steady, then fell quickly again as Romanians adapted. Because an unplanned pregnancy in 1968 could simply move a birth earlier without changing a woman's eventual family size, the authors caution (citing Lawson and Spears, 2025, on the theoretically ambiguous effects of abortion restrictions) that period-rate jumps can reflect timing shifts rather than more total children. Panel B, plotting completed cohort fertility by the year each cohort turned 30, confirms this: Romania started higher than Bulgaria but showed no sustained additional rise relative to Bulgaria during the decree's enforcement. Tyler Cowen notes Hungary's subsidy-driven baby boom has now reversed into a bust, possibly via the same timing mechanism.

Fertility has fallen worldwide because birth control got much better and easier to use, and because people — women especially — simply do not want many children. The pill spread through wealthier countries in the 1960s-70s, but diffusion is slow: norms need generational turnover, and "fertility contagion effects" take a long time to propagate. Social science has other long-lag mechanisms, e.g. how long it took migrants to exploit open borders before WWI, and wealthy-country fertility has been declining for a long time. (Richard Hanania is cited for arguing income isn't the real driver and the decline keeps continuing.)

The result: many women, facing the real stress and cost of parenting, want only one child, producing large numbers of one-child families, some who never marry or have kids, and a smaller share with 2-4 kids — plus cases where the father leaves, self-destructs, or never marries after one child. Many women love having a child enough that one is fully satisfying. Cowen compares this to owning his dog Spinoza: no desire for a second dog, since "dog attention" supply is elastic; a single kid similarly absorbs much of a parent's time and affection. Because the model is preference-driven, fertility decline should be very hard to reverse — matching the evidence so far, and a genuine problem (less well-being, less innovation, possible fiscal crises).

The FDA under the new administration hasn't become less paternalistic, just differently so — fixated on curbing mRNA vaccines under RFK Jr. It issued a rare Refusal-to-File letter to Moderna for its mRNA flu vaccine, mRNA-1010, rejecting an application built on Phase 3 trials of 43,000+ participants costing hundreds of millions under FDA-agreed design. The stated reason: Moderna's control arm used a standard flu vaccine rather than an ACIP-preferred "enhanced" one for adults 65+. Cowen argues that's defensible alone but ignores the value of a diverse "medical armamentarium" — mRNA updates faster and can outperform in practice despite lower trial efficacy — and that retroactively changing agreed rules is a textbook case of regulatory uncertainty destroying option value. The FDA had approved this design in April 2024 and reaffirmed it in August 2025; instead Vinay Prasad overrode staff ready to proceed, despite Moderna supplying the requested data — Moderna unusually publicized his letter. The EU, Canada, and Australia accepted the application. Meanwhile HHS canceled $500 million in BARDA contracts and a $590 million avian-flu contract, states are pushing mRNA bans, and Moderna's CEO says the company won't fund new Phase 3 trials without US market access — EBV, herpes, and shingles vaccines are already shelved. Cowen calls it medical nationalism, not medical freedom, risking export of a general-purpose platform with spillovers into cancer vaccines.

Montana's new right-to-try law, SB535, could seed an American biotech comeback by letting states do what the FDA won't. China is winning the drug race through deregulation: in 2024 its NMPA approved 83 new drugs versus the FDA's 50, its share of global commercial clinical trials rose from 8% in 2013 to 30% in 2024 (just behind the US's 35%), and Jiangsu Hengrui Pharmaceuticals overtook AstraZeneca as the world's top trial sponsor. Hainan's Medical Tourism Pilot Zone even lets clinics use any drug already approved in the US, EU, or Japan without separate Chinese review. The US system, by contrast, demands billions of dollars and a decade-plus of trials before any revenue, causing drug lag, drug loss, and lost innovation momentum. Signed in May 2025, SB535 lets Montana-licensed "experimental treatment centers" sell drugs that have merely cleared Phase I, enabling a self-funding pipeline toward full FDA approval; these cash-only treatments (no Medicaid, Medicare, or likely private insurance) target patients who've exhausted conventional options. Florida's SB 1768 similarly permits non-FDA-approved stem cell therapies for orthopedic, wound, and pain conditions. The author urges HHS and the FDA to cooperate with these state frameworks rather than obstruct them.

HHS's new Operation TrialBlazer aims to speed US clinical trials, explicitly motivated by China's rise: China's Phase 1 trial share overtook the US's in 2021, and in 2024 China registered over 7,100 trials, 39% of the global total. TrialBlazer's substantive reform is clarifying CMC (Chemistry, Manufacturing, Controls) requirements, since the FDA had forced companies to prove years-long formula stability and document full commercial manufacturing processes for drugs still in early testing—costly, low-value overkill now being trimmed.

But TrialBlazer avoids the two models it cites as exemplars. China lets investigator-initiated trials proceed for cell, gene, radioligand, and stem-cell therapies whenever a researcher has funding and interest, paired with aggressive industrial policy to back winners. Australia's Clinical Trial Notification system requires no prospective government review of most early-phase trials: sponsors get ethics-committee (HREC) approval, then merely notify and pay the regulator (TGA), yielding approvals in 21-28 days and starts under 70 days—though the TGA still vets the highest-risk biologicals. Australia's model, which "certifies the certifiers," was proposed by ex-FDA official Henry Miller and long favored by the author. Instead, the US keeps the FDA as full gatekeeper, merely making that gate faster and cheaper to clear.

As populations shrink and AI absorbs most informational functions, each remaining human will become individually noticeable, pushing people to compete on appearance and charisma rather than information. Tyler Cowen argues emptier public spaces (a fertility-crash effect) mean anyone encountered in person draws more attention — sidewalk chat and small-town-style greetings will spread even to big cities, which will feel like smaller towns due to lower density. Since AIs already handle "useful informational functions," anyone wanting to stand out must specialize in what's left to humans: touching grass, giving warm hugs, looking good or interesting, and projecting a unique identity that AI smart glasses can communicate during encounters. YouTube figure "Clavicular," mocked for extreme "looksmaxxing" (heavy appearance manipulation plus plastic surgery) despite no other accomplishment, is cited as an early harbinger. Expect more investment in dress, makeup, height, and weight, with plastic surgery and successor drugs to GLP-1s commanding even greater interest. Nondescript people risk being ignored in favor of chatting with an AI instead; ordinary social life starts resembling a gala where everyone dresses to be noticed. Cowen admits this sounds ridiculous now but notes people already care far more about looks than medieval humans did, and adapted.

Zephyr Teachout's New York Times op-ed on grocery prices is economically incoherent, built on claims that contradict each other and misread ordinary market behavior. Teachout argues the food system is rigged because big retailers extract special deals from suppliers, who then "make up the difference" by charging smaller stores more. Cowen counters that volume discounting (Costco offering suppliers lower prices for bigger orders) isn't anti-competitive, and firms actually compete to sell to Costco for its sales volume and brand exposure — over a quarter of Americans pay for a Costco membership. Profit-maximizing suppliers already charge what the market bears elsewhere; a discount to one buyer doesn't force higher prices on others.

Teachout's egg example backfires: a dozen eggs cost $5.99 at an independent store versus $3.99 at a major retailer nearby, yet she blames "market power" — meaning the accused abuser is the one charging less. She also invokes NYC's price-gouging law, which exempts cost-driven increases, then cites bird-flu-driven egg price hikes (a textbook supply shock) as the abuse to police. Finally, within a few paragraphs she accuses big retailers both of pricing unfairly low and unfairly high — missing only an accusation of collusion to complete the contradiction.

The US Constitution makes corporate nationalization essentially impossible, shaping a decentralized American corporatism instead. The Fifth Amendment's takings clause requires compensation for seized property, and since America never built up state-owned enterprises, there is no cheap legal path to nationalize -- it would be costly to government itself, and strong corporate interests remove pressure to override the amendment. Extreme federalism reinforces this: overlapping federal, state, and local regulators diffuse control in ways that preempt the centralization nationalization requires. The pattern dates to the mid-to-late 19th century, when a still-small federal government (local government then accounted for roughly half of public spending) let giants like the railroads, Bell, and Western Union grow large before Washington matured, pushing regulation to the state level instead (Millward 2013). The tradeoff: strong property rights, but weak infrastructure-building, since the same legal checks enable NIMBYism and litigation by any interested party -- illustrated by a Port Authority review inviting the Shawnee Tribe of Oklahoma and Sand Hills Nation of Nebraska (Howard 2014). Because national consensus sustains political support, government doesn't need state firms to reward insiders, though crony capitalism persists. The US instead runs stable, bipartisan industrial policy: trade deals and military-alliance leverage keep entertainment markets open abroad regardless of party.

Robin Hanson and Vitalik Buterin deserve a Nobel Prize in economics for applied mechanism design, now that prediction markets have gone mainstream. The trigger: Intercontinental Exchange (owner of the NYSE) announced a $2 billion investment in Polymarket, the Ethereum-based prediction-markets platform, with NYSE planning to use Polymarket data to sharpen its forecasts. Polymarket founder Shayne Coplan credits Hanson's writing on prediction markets as his original inspiration, recalling that he started the company at 21, broke, 2.5 years after dropping out, at the onset of the pandemic.

Cowen argues the next step is decision markets, not just prediction markets. As an example, Futarchy just launched a market on Tesla's share value conditional on whether Musk's pay package is approved. Scaled up, this resembles Hanson's 1996 "Dump the CEO Market" proposal: a market pricing a company's shares with versus without its current CEO. Cowen recalls featuring Hanson's Decision Markets paper in his 2002 book "Entrepreneurial Economics," where he predicted newspapers would one day cite market prices on policy questions (e.g., "health care plan A") instead of quoting experts — a moment he says has now arrived. An addendum links Cowen's a16z podcast with Scott Duke Kominers on prediction markets.

Polymarket's prediction markets are remarkably well-calibrated, rivaling state-of-the-art weather forecasting in accuracy. Dune analyst Alex McCullough measured this using the Brier score — the mean squared error between a market's stated probability and the actual binary outcome (a 70% prediction that resolves "yes" scores (0.7−1)²=0.09; a 90% prediction on the same event scores a better 0.01; lower is always better). Across roughly 90,000 predictions, Polymarket's 12-hour-ahead Brier score is 0.0581 — well under the 0.125 "good" and 0.1 "great" thresholds, on par with the best forecasting models and far better than sports betting lines (0.18–0.22). That puts it in the range of elite 12-hour rain forecasts (0.05–0.12). Liquidity sharpens accuracy further: markets with over $1 million in trading volume score 0.0256 twelve hours out and improve to 0.0159 a day before resolution, though even thin markets stay under 0.1. One systematic flaw persists: a chart bucketing predictions into 20 five-percent bands shows predicted probabilities track actual outcome frequencies closely and progressively, but predicted odds run consistently a touch high — traders slightly overpay, possibly from shorting frictions, automated-market-maker mechanics, or favorite-longshot bias.

The U.S. should privatize air traffic control by copying Canada's 1996 model, since the real problem is bad incentives, not just staffing. The trigger: the FAA grounded flights at Reagan National and delayed thousands nationwide, short roughly 3,500 controllers and forcing mandatory six-day overtime, sometimes unpaid during the shutdown. The deeper flaw is that a mission-critical service runs on annual congressional appropriations subject to political failure. Canada fixed this in 1996 by spinning off air navigation into NAV CANADA, a private non-profit funded by user fees rather than taxes; safety regulation stayed with government while operations moved to a professionally governed, bond-financed utility on multi-year budgets. NAV Canada became the first air navigation provider worldwide to deploy space-based ADS-B, with the UK's NATS, over the North Atlantic in 2019, and the first to use it domestically—giving Canada satellite-based rather than ground-radar navigation. By contrast, an Inspector General report found the FAA's NextGen modernization, over 20-plus years of delays and cost growth, had delivered only about 16 percent of its expected benefits as of December 2024. The fix: fund U.S. air traffic control with airline user fees and seat airline executives on the governing board, aligning power with incentives so performance follows.

Australia's ban on under-16s from platforms like YouTube and X will do more harm than good. Targeting YouTube especially will hurt youth science education and young founders' ability to launch projects; a country that only lets people learn from the internet at 16 will produce fewer great achievers at 18 or 20. More seriously, the law expands government control over speech: officials must arbitrarily decide which services count, and loopholes already show — gaming (Fortnite) and messaging (WhatsApp, Discord) are exempt, so social behavior will migrate there, forcing the ban to keep expanding, with AI regulation a likely next target. Cowen predicts either weak enforcement or ever-tightening controls; in a large poll, only 6% of Australian 9-to-16-year-olds thought the ban would work. He asks why the state, not parents, should stop his 15-year-old from following chess champion Magnus Carlsen on X and YouTube, comparing the move to Socrates's Republic, where the state supplants the family in choosing children's stories. Since Australia lacks First Amendment-style protections and already restricts "hate speech," he doubts regulators can be trusted with more power. He grants that phone-use limits show small real gains, but argues this law's benefits will be modest against its cost to free speech, noting Ben Yeoh's companion piece surveys the research on social media and teen harm.

Lunar settlement will follow the same sequence as New York's 1811 street grid: impose legible order first, let private development follow. Charles Miller, space entrepreneur and head of the Trump NASA transition team, proposes a Lunar Development Authority (LDA) in a CSIS piece — a US-chartered quasi-governmental regulator treating the Moon as a master-planned infrastructure project, with NASA as strategic partner, the US and allied governments as anchor tenants, and public-private partnerships financing construction. The model is Manhattan's Commissioners' Plan of 1811, which gridded undeveloped land with standardized lots before development arrived. Miller's sequence: survey, standards, shared infrastructure, and governance first; private tenants, resource extraction, and finance after.

The main obstacle is the 1967 Outer Space Treaty, which bans WMDs and national land appropriation but not commercial activity. The Artemis Accords clarify that resource extraction is compatible with the OST. Unlike the Homesteading Act, which granted title in exchange for development, the lunar path reverses that order: resource ownership (already accepted) comes first, followed by exclusive operational zones, long-duration concessions, and eventually transferable development rights around fixed infrastructure. The OST can delay formal land markets but not the underlying economic requirement that builders retain enough of what they create to make settlement worthwhile.

The Davis-Bacon Act, which forces federally funded construction projects to pay "prevailing wages," should be rescinded because its costs go far beyond a simple wage transfer. Drawing on an account by Mike Schmidt, Director of the CHIPS Program Office, Cowen argues the law is a major source of friction, not just expense. It isn't a flat minimum: rates vary by trade and locality, requiring the Department of Labor to maintain over 130,000 separate wage rates. Applying it to firms' own salaried employees (rather than contractors) proved especially disruptive — hours had to be tracked by trade classification, weekly-pay mandates clashed with profit-sharing compensation, and internal strife emerged when equally-paid staff were split based on who did "construction" work. Because 1930s trade categories don't map onto semiconductor fabs, applicants and contractors navigated classification largely from scratch. Worst was retroactivity: firms encouraged to break ground before rules were finalized later had to reconstruct back pay for up to 20,000 already-departed workers across 300-plus subcontractors, adding hundreds of millions in cost. Since the work and pay were already settled, this was a pure windfall to a politically favored group, not an incentive fix. Compliance costs likely rivaled the wage premiums themselves — a very leaky bucket.

The Southern Poverty Law Center did commit bank fraud, and it's not a close call, even though the underlying conduct was an open secret. SPLC set up fake businesses to pay secret informants, then lied to a bank about who owned and controlled those accounts—a violation of 18 U.S.C. §1014, since almost any false statement to a federally insured bank is illegal and banks act as an investigatory arm of the state. When Bank-1 investigated, an SPLC employee asked it to close the accounts and move balances to an SPLC account; the president/CEO and board chair then confirmed in writing that the accounts were SPLC-operated—"a succinct confession to bank fraud." Yet the informant program had been public knowledge for decades, raising the question of whether motive should matter (legally, it doesn't).

The backstory: SPLC's blacklist became embedded in private-sector systems—Amazon used it for AmazonSmile, workplace-giving vendors relied on it—giving the linked Change the Terms coalition leverage over tech and finance firms. This was tolerated while targets were groups like the KKK, but the coalition later went after Trump-aligned organizations, holding hundreds of meetings with top-level staff. A Free Press fundraising letter bragged of pressuring Twitter to ban Trump and pushing Facebook toward a permanent ban, while claiming nonpartisanship. Trump won; the indictment followed—politically motivated, perhaps, but not legally weak.

American science has long been state-funded but not state-directed—since Vannevar Bush, money has flowed through many agencies to independent universities via peer review—and a proposed OMB rule, joined by some forty agencies (NSF, HHS, DOE, NASA, DOD), rewriting the "Regulation for Federal Financial Assistance," would end that by bringing funding under top-down political control. Program goals must align with "administration policies and priorities"; senior appointees, not peer reviewers, must ensure awards advance "the President's policy priorities," with peer review merely advisory; every grant becomes terminable whenever it no longer serves "the national interest as they exist at the time"; and subrecipients must not damage the federal government's reputation—a loyalty clause. Though sold as cutting burdensome conditions, the rule actually adds bureaucracy: written payment justifications, Treasury "Do Not Pay" screening, mandatory E-Verify, disclosure of recently agency-affiliated employees, a new pre-issuance review committee, and elimination of output-based "fixed amount awards" for routine cost monitoring. OMB's guidance becomes binding government-wide with no agency rulemaking—one White House dial controlling every grant program. The drafters admit the politicization, framing it as retaliation for Biden-era equity mandates (E.O. 13985). Decentralized funding is inefficient like markets or federalism, trading scale for experimentation and robustness; centralizing it mimics ministries of science—an experiment whose outcome is already known.

About a quarter of the US workforce now needs a license to work, up from 5 percent in 1950 — but in 1983 Colorado delicensed funeral directors, an unusual reversal of that trend. Tyler Cowen and Brandon Pizzola's 2017 paper on this natural experiment found delicensing lowered wages and prices and shifted consumers toward cheaper cremation over the mortuary services directors preferred. A chart of average weekly wages, Colorado versus the rest of the US, shows the two tracking together before 1983 then diverging, with Colorado's wages falling below the licensed national average.

In 2023, revelations of body-part sales and decomposing bodies at a Colorado funeral home — blamed on Colorado being the only state without licensing — led it to relicense the industry starting 2027. Cowen argues this is a mistake: comparably gruesome abuses, including sexual abuse of corpses, occur in heavily licensed states too. Colorado's own 1990 sunrise review and a 2015 CEA report both found no evidence licensing improved safety or that its absence caused harm. Training-hour requirements don't target the actual abuses, and the offending director already got 40 years in prison — deterrence enough. People want guaranteed good behavior, and licensing gets reached for even when it can't deliver that.

Prediction markets already reveal useful information, but Robin Hanson's "decision markets" would do it far better by pricing conditional bets directly instead of forcing us to infer them. Tyler Cowen, via a pointer from Arin Dube, illustrates with a chart of a betting market: a scandal sent the probability that Graham Platner drops out of the Maine Democratic primary from 9% to 96% (an 87-point jump), while the probability Democrats win the general election rose from 54% to 63% (9 points). Since an 87-point rise in dropout odds produced a 9-point rise in win odds, a full (100%) chance of dropping out implies roughly 9/0.87 ≈ 10.3 points of gain — meaning Platner is costing Democrats about 10 points versus an unknown replacement. But this reverse-engineering assumes the news moved only the dropout probability, when really it's trying to back out two conditional prices, P(win|drop) and P(win|stay), from one unconditional price. Hanson's decision markets would instead trade contracts paying $1 if Democrats win conditional on Platner dropping out (refunded otherwise), plus the mirror contract — yielding clean, continuous conditional estimates without needing a scandal to reveal them. Prediction markets are catching on, Cowen notes, but the world remains decades from decision markets, let alone Hanson's futarchy.

The Trump administration's rewrite of the Regulation for Federal Financial Assistance threatens America's historically decentralized, politically insulated system of science funding, as Tyler Cowen warned earlier and others now confirm. The AAAS calls the rule a "brazen power grab" by the OMB director, warning that an Alzheimer's cure would come to depend on "the scientific sensibilities of the nation's chief bureaucrat." At The Dispatch, Grayson Logue writes the rule would let the president and appointees cancel funding or target recipients for almost any reason; the White House calls this accountability, but critics fear politicization and corruption from an administration already accused of punishing enemies and rewarding allies. Dan Drezner calls it exactly the rule he'd write "if trying to ruin American leadership in scientific research," part of a broader assault on state capacity and impartiality. Noah Smith says MAGA's attack on science is worse than it looks, pairing funding cuts with ideological review by "government commissars" despite science's broad public trust. Cowen's own worry: funding was never fully apolitical, but decentralization kept culture-war intrusion marginal; COVID policy, gender policy, and now this rule's weaponization of those fights may have ended that equilibrium. Like judicial independence, decentralized science is a fragile American exception to a global default where centralized control simply serves whoever holds power.

Government pay relative to GDP per capita rises the poorer the country, meaning public workers are most overpaid where economies can least afford it—and this excess distorts labor markets on two margins: rent-seeking queues for jobs and misallocation of talent away from the private sector. Cowen's earlier posts on India covered the queue margin: educated youth burn years cramming for exams that teach no real skills, entire towns specialize in exam prep, and India ends up with roughly one-fifth the civil servants per capita of a rich country despite rich-country laws—rent-seeking losses he estimates near 1.4% of GDP annually. Two macro papers quantify the misallocation margin. Geromichalos and Kospentaris, modeling Greece, find a 10% cut in public wages would raise private-sector productivity 3.8%, cut unemployment 7.3%, and lift GDP 1.3%; their model assumes perfect sorting, so no rent-seeking waste exists. Cavalcanti and Santos, modeling Brazil, find narrowing the public-private wage premium from 19% to 15% and aligning pensions could raise output 11.2% long-run without cutting public services—and separately estimate rent-seeking costs at 3.61% of output, triple Cowen's India figure. Across all three countries, overpaid government work drains private-sector talent and productivity, making high relative government pay a development trap for poor economies.

Singapore's famously high pay for top officials does not contradict the claim that high government pay wastes talent in poor countries, because Singapore runs on market wages, not the rents that plague India, Greece, and Brazil. Answering a reader who cited Singapore against his earlier argument, Tyler Cowen notes the countries aren't comparable: Singapore's GDP per capita is 37 times India's (~$91k vs. $2.4k), with 1/233rd the population (6m vs. 1.4b).

The critique targets millions of ordinary civil servants, not a few ministers. Justin Sandefur's data show teacher pay relative to GDP per capita is highest in the poorest countries, sometimes five times GDP per capita — well above private-sector wages for equivalent work, what Sandefur calls "rents." Singapore breaks this pattern: teachers earn just 70-80% of GDP per capita, ordinary market wages. Only its top tier is unusual — fewer than 500 officials pegged to the top 1,000 citizen earners, correcting for the wage compression that squeezes elite pay in unionized democracies like Denmark.

Paying market rates throughout lets Singapore skip the mass-exam gatekeeping poor countries need — unlike Italy, where 85,000 applicants (shortlisted to 8,000) competed for 30 low-glamour Bank of Italy jobs.

Capitalism and democracy are not enemies but mutually reinforcing, contrary to two rival views: that democracy kills capitalism (classical economists from Smith to Mill, Marx and Engels welcoming that outcome, and today Peter Thiel arguing freedom and democracy are no longer compatible) and that capitalism kills democracy (Robert Reich and Joseph Stiglitz, citing corporate lobbying and bribery). A third tradition — Hayek, Mises, Friedman — held the two go together; Mises wrote liberalism "must necessarily demand democracy." A new paper by Tyler Cowen and Vincent Geloso, in the book "Can Democracy and Capitalism be Reconciled?," backs this empirically. Plotting Fraser Institute economic-freedom scores against V-Dem democracy scores shows every major democracy is capitalist and nearly every capitalist country democratic (exceptions: Singapore, Hong Kong); the high-democracy/low-freedom "democratic socialism" quadrant is empty. The Berlin Wall's fall shows a democracy shock driving sustained later freedom gains. Case studies of the biggest authoritarian-era freedom jumps (Peru, Nicaragua, Uganda, Chile) show authoritarian rule can also crush freedom — Peru's fell under 1970s authoritarianism before Fujimori's later gains were ratified, not reversed, once democracy returned. Two mechanisms: only growth, not redistribution, enriches a majority, giving democratic rulers incentive toward capitalism; and centrally planned economies (Robert Dahl) hand leaders resources that corrupt, collapsing democratic socialism into authoritarianism.

The False Claims Act, which lets whistleblowers sue defrauding firms on the government's behalf for up to 30% of recovered funds, generates enormous deterrence value beyond the money it directly recovers. Economist Jetson Leder-Luis studies Medicare fraud, where the government relies on providers' self-reported billing, making insider tips valuable. His central case is Tenet Healthcare, sued in 2002 for manipulating cost reports to inflate "outlier" payments (for unusually expensive patients) and settling in 2006 for $788 million. Outlier payments, running about $500 million a month, fell by more than half after the suit—even at hospitals never sued—implying industry-wide deterrence.

Leder-Luis estimates the five-year discounted deterrence from that case alone at $17.46 billion, roughly nineteen times the $923 million in outlier settlements. Across four case studies where whistleblowers recovered $1.9 billion, he estimates $18.9 billion in deterrence effects, dwarfing the program's costs: 2018 lawsuits cost the public under $108.5 million, and total payouts since 1986 total $4.29 billion. Cowen notes skepticism of Leder-Luis's synthetic-control method (extra free parameters) but, given the scale of gains, endorses the conclusion: privatized enforcement, driven by profit motive, is a highly effective, low-cost fraud deterrent.

American democratic institutions are functioning normally despite widespread elite fears of collapse, though corruption remains a real problem. Tyler Cowen cites evidence: Democrats who won November elections will take office without incident; the Supreme Court looks likely to strike down major parts of Trump's tariff plan, and Trump has already reversed some food tariffs over "affordability"; National Guard troops were pulled from Chicago and Portland partly via court challenges; the D.C. deployment proved a civil-liberties non-event. He lists events from November 18: the House voted 427-1 to release the Epstein files; a federal judge blocked the GOP's Texas redistricting map, letting Democrats net seats for 2026; a federal appeals court (including two Trump appointees) rejected Trump's defamation suit against CNN over "Big Lie"; Corporate Public Broadcasting agreed to honor its $36 million NPR contract after a judge found its defense not credible; and a New York judge dismissed Trump's challenge to a state law barring immigration arrests in courthouses. The Senate followed suit on Epstein, and commentators like Nate Silver now call Trump a possible lame duck. Cowen argues the "democracy is disappearing" fear was a major elite error, driven by anti-Trump emotion, pessimism bias, recency bias from an early quiescent Congress, discomfort admitting democracy itself produces bad outcomes, and a naive "good people will good things" theory of politics lacking any sense of spontaneous order.

Private property, not collective ownership, saved the Plymouth colony from starvation, and Governor William Bradford's own account of why is one of history's clearest statements of political economy. Cowen re-ups his 2004 post: the Pilgrims first organized as a communist society and were soon starving. "After much debate of things," Bradford ended the shared-corn system, letting each family keep what it grew. Quoting Bradford, the piece shows the change "made all hands very industrious," yielding far more corn than the Governor could ever have compelled, with women and children now going willingly into the fields. Under the old common system, able young men resented laboring for others' families without extra reward, the strong got no more food than the weak, elders felt indignity at being leveled with youths, and wives forced to serve other households called it a kind of slavery — collectivism eroded "mutual respects" even among "godly and sober men." Bradford preempted the objection that this was mere human corruption, not the system's fault: since all men share that corruption, God intended private property as the fitter course. Cowen notes Bradford's claim that enforcing equality would have required "great tyranny and oppression" — a lesson that could have spared the twentieth, and per his 2025 addendum the twenty-first, century much pain.

Mass incarceration in America was driven primarily by violent crime, not racism — and black voters and leaders were themselves among its strongest supporters.

Tyler Cowen revisits a point from the Marginal Revolution Podcast on "Crime in the 1970s": against Michelle Alexander's The New Jim Crow, which frames mass incarceration as a racist project, Alex Tabarrok noted the era's justice system was also called racist, but for the opposite reason — black criminals were released back onto the streets to terrorize black victims and given sentences seen as too light. Black and white victims together pushed for tougher punishment. This undercuts both the claim that racism primarily drove mass imprisonment and the claim that black leaders don't care about black-on-black crime; many were and remained tough-on-crime through the 1970s-80s crime wave, with misgivings surfacing only after crime fell in the 1990s.

A new paper, "The Racial Politics of Mass Incarceration" by Clegg and Usmani (Perspectives on Politics), supports this: both white and black public opinion turned more punitive after the 1960s, most black House members voted punitively at crime's peak, and post-1990s-redistricting black political representation had a punitive effect at the state level. Racism still shaped how crime fears became policy, but crime, not racism, was the primary driver.

Official complaint statistics vastly overstate genuine public grievance, because a handful of serial complainers generate most of the volume that institutions must investigate at taxpayer expense. Dourado and Russell found that in 2015, 6,852 of 8,760 noise complaints to Reagan National Airport came from a single Foxhall (DC) residence, whose occupants called nearly 19 times a day. By 2024 total complaints had exploded but stayed concentrated: one person filed 20,089 complaints (25% of all), while just 188 total complainants logged 79,918 complaints, averaging 425 each (over one per day).

The same pattern appears in Department of Education Office for Civil Rights data: in 2023 one individual filed 5,059 of 8,151 sexual-discrimination complaints (68.5%). Recent OCR reports name the complaint type behind repeat filers, unlike earlier years, letting Cowen chart totals against single-individual volume—likely an undercount, since other categories like age discrimination show similar filer concentration. The accompanying graph shows single filers repeatedly accounting for 10-30% of all complaints across years.

Cowen likens this to one person repeatedly pulling a fire alarm without consequence, and asks whether Congress and media adjust reported totals for such spam or simply repeat them as-is—warning that institutions increasingly exist to manage a tiny number of neurotic, possibly malicious complainers.

Principled resignation is an adverse selection: when the ethical quit and the unethical stay, the institution rots — so Tyler Cowen argues resigning is often the wrong call, prompted by a wave of DOJ resignations (six prosecutors and an FBI agent) over pressure to investigate the widow of Renee Good, killed by ICE agent Jonathan Ross, rather than the agent, plus the Alex Pretti killing and apparent cover-up.

Resignation can still work as a signal and can create common knowledge, as when Attorney General Elliot Richardson resigned rather than fire Watergate special prosecutor Archibald Cox — but Cowen asks who plays that role today. The case for staying is retaining "voice" (citing Hirschman): forcing written directives, triggering inspector-general review, escalating through professional-responsibility channels, building coalitions. The case against is that staying corrupts — small compromises accumulate until intolerable lines feel normal, "banality deadens evil." Resignation keeps hands clean, which utilitarians call moral self-indulgence, but Bernard Williams argues genuine moral commitments sometimes require it.

Cowen offers four decision factors — complicity, voice, timing, self-discipline — and concludes that right now, voice is more socially effective than exit.

Guarantees that shield one group's income from economic tradeoffs breed resentment rather than gratitude, because voters can demand contradictory things at no individual cost. The UK's Triple Lock raises pensions by whichever is highest of wage growth, inflation, or 2.5 percent -- "the single stupidest policy in the entire Western world," per Luis Garicano, a verdict Tyler Cowen endorses. The mechanism bites hardest when the economy is weakest: during the 2009 crisis, wages fell and inflation dropped, yet pensioner incomes still rose 2.5 percent. Matt Yglesias finds the same pattern in the US, where per-capita elderly benefits have grown steadily more generous even as the ratio of retirees to workers has risen -- driven by broad public support and bipartisan elite acquiescence -- yet voters stay angry about stagnant public services, blaming fraud (right) or undertaxed billionaires (left) rather than the real driver: transfers to seniors crowding out everything else. Cowen calls this Bryan Caplan's "rational irrationality": individuals pay no price for wanting generous pensions, robust services, and low taxes simultaneously, since no single vote matters -- compounded by Arrow's impossibility theorem, whereby even coherent individual preferences can't aggregate into coherent collective choice, so contradictions surface only when the bill arrives.

Baltimore's post-Freddie Gray collapse in policing didn't stay temporary — it tipped the city into a permanently higher-crime equilibrium, exactly as warned a year earlier: a crime spike depresses arrest and clearance rates, which in turn sustains the higher crime rate, a self-reinforcing multiple-equilibria trap. Writing at FEE, Daniel Bier reviews Baltimore's history and confirms the mechanism: a chart of 2015 arrest data shows the sharp drop in arrests was indeed short-lived, yet once police pulled back they proved unable to cope with the new normal, leaving Baltimore's crime cycle stuck in high gear. With 178 killings already logged in 2016, the city is on pace for roughly 294 murders — below 2015's 344 but well above 2014's 211; scaled to New York's population, per a line from HBO's The Wire, that would translate to nearly 4,000 murders a year. Bier argues Baltimore has the resources to recover but may lack the will. The city is worse positioned than Ferguson, MO, which had poor policing but only an average crime rate; Baltimore combines poor, distrusted policing (partly justified, per the DOJ's report) with sky-high crime. A federal funding "surge" could fix both, but risks being denounced as a takeover — one a law-and-order administration might impose yet be less likely to make succeed.

Baltimore's murder collapse is best explained not by the national crime decline alone but by a targeted program combining predictive intervention, services, and deterrence. Tyler Cowen recalls predicting in 2015 that Baltimore would tip into a high-crime equilibrium after the Freddie Gray riots, since a crime surge strains police resources and that strain itself lowers the odds of punishment, sustaining high crime even once arrests rebounded. Yet since around 2022, murders have fallen sharply: April 2026 saw just four homicides, the lowest month since at least 1970; year-to-date the city has had 38, versus 51 over the same period last year and 323 four years ago — putting 2026 on pace for under 100. Charles Fain Lehman's Free Press piece attributes this to murder's non-random structure: a killing is often followed by predictable retaliation, usually carried out by a close associate of the victim. Under Mayor Brandon Scott and a tough prosecutor, Baltimore adopted a focused-deterrence program modeled on Boston's Operation Ceasefire, weekly reviewing every shooting and dispatching outreach teams to victims' known associates. Those contacted are offered "carrots" — job training, tattoo removal, relocation — alongside an explicit "stick": a mayoral letter warning they are being watched.

Even highly capable AI will do little to lower the cost of living for ordinary Americans over the next five years, because the constraints binding most household budget categories are supply-side, institutional, and legal rather than technological. Tyler Cowen tests this against four major expenditure categories: rent, food, health care, and education.

Rent is mostly a supply problem AI can't fix; strong AI could even raise rents by making it more valuable to live near talented people, though it might also make remote locations like Maine more viable, lowering rents de facto. Food could get cheaper through AI-driven crop engineering, irrigation, weather prediction, and land-trading systems, but such gains diffuse slowly through real-world institutions -- much of rural Mexico still lacks tractors -- so a twenty-year horizon is more realistic, and in the short run rising energy demand could make food costlier. Health care will see genuine breakthroughs, potentially pushing life expectancy toward 97, but longer lives and more (sometimes customized, cancer-treatment-style) interventions mean total health spending is likely to rise, not fall, even as welfare improves. Education is already cheaper and more effective for the motivated, though largely by freeing up leisure time rather than cutting costs, while lower-grade learning stays unchanged and credentialing costs persist.

Cowen concludes AI could be superb in intrinsic quality while living costs stay flat or rise, meaning AI may prove less popular with the public than its technical merits would suggest.

AI industry's circular deals — companies investing in, buying from, and selling to each other in loops — resemble ordinary vendor financing, not fraudulent revenue inflation, and don't warrant special alarm. Tyler Cowen endorses Noah Smith's analysis, which identifies two fears about such deals: that they artificially inflate revenue to fool investors, and that they raise systemic risk by tying AI firms' fortunes together. On the first, Smith distinguishes "round-tripping" from vendor finance using a hypothetical: startups Aegnor and Beleg secretly agree to buy ad space from each other, inflating both companies' revenue with no real profit or cash flow — illegal collusion that occurred during the dotcom boom, and can happen unintentionally, without collusion, whenever many startups sell mainly to each other. Vendor finance, by contrast — like GM Financial lending customers money to buy GM cars — is legal and healthy because the underlying transaction has genuine value. Smith judges AI's deals closer to the latter. Cowen agrees, adding that the deals' terms are widely publicized, so no serious investor is being deceived, and draws a parallel to horizontal/vertical integration, which also carries a financing element; since no corporate control changes hands in these AI deals, mutual investment alone isn't inherently suspicious.

Wealth taxes are the wrong tool for capturing AI-era gains; more targeted taxes already fit the job. Responding to Dwarkesh Patel's tweet on his paper with Trammell, Tyler Cowen proposes taxing valuable homes in good locations as a "housing services tax" rather than a wealth tax, and raising consumption taxes on luxury goods like paintings and yachts. He cites economist Tom Holden's argument that once capital becomes essentially the only input to production, taxing it slows growth (versus today's mere level effects), so capital taxes should grow less attractive, not more, as labor's income share falls. Cowen expects health-care spending to rise sharply as AI extends lifespans, making health-care taxes a natural fiscal lever—accepting shorter lives (95 versus 97) as a tradeoff, with savings redirected toward birth subsidies. He also cautions that capital won't fully substitute for labor anytime soon, and that final incidence depends on supply-and-demand elasticities, not just substitutability—cheaper AI-driven services could raise real wages rather than lower them. Capital "having all the goodies" isn't an equilibrium: even a hypothetical monopolist producing all output with one watt of energy would need to cut prices to sell it. Dwarkesh's follow-up, Cowen notes, envisions a far more distant future involving ownership of galaxies.

There is no aggregate demand problem in a rapidly advancing AGI world. Even amid radical deflation tempting people to hoard money, the high marginal utility of incredible new goods pulls spending forward, since future marginal utility only falls amid plenitude — a Pigou wealth effect robust even to losing one's job. Fed policy still matters, since others will use fresh liquidity to invest, including in AI goods not all controlled by "billionaires." Liquidity-trap arguments need a genuinely miserable investment and consumption environment; besides, liquidity traps were meant to apply only to currency, and extended to money broadly, where most holdings bear interest, the concept collapses. Say's Law holds: AI-generated output creates income, and prices adjust (falling further if unemployment rises), so abundant goods plus billions accruing to AI owners cannot coexist without prices settling where people can afford to buy — revenue can't stay trapped in currency hoards (a Huttian/Walrasian point, citing W.H. Hutt). The real risk is regional: wealth concentrating near energy and data centers, forcing migration reminiscent of the Rust Belt — worrying, but not collapse-level. The "Citrini" 2028 doom scenario is dismissed as unsound macroeconomics, per Eli Dourado's and Nicholas's critiques.

AI-accelerated drug discovery will collide with FDA bottlenecks, and the nearer-term gain lies in repurposing, not new approvals. Tyler Cowen builds on Dean Ball's warning that AI will soon generate batches of computationally validated drugs while FDA backlogs keep Americans from benefiting, possibly pushing pharma startups toward friendlier jurisdictions like the UAE, until regulators eventually break the logjam in ways seen as reckless and politically charged. Cowen argues the faster payoff is off-label prescribing: once a drug is approved for one use, physicians can apply new evidence to any use immediately, skipping fresh FDA trials. When the RECOVERY trial confirmed already-approved dexamethasone helped some COVID patients, doctors prescribed it within hours—requiring new efficacy trials could have cost a million more lives. With thousands of drugs already approved, AI can mine side effects as candidate main effects for other conditions.

Cowen sees three ways AI-discovered drugs will strain the FDA: cheaper, more numerous candidates make rival regulators—China now produces more new drugs than Europe, per an accompanying chart—competitive; AI enables "rational drug design," diagnosing and fixing problems the way a mechanic does rather than running RCTs; and personalized medicine (breast cancer has split into ten molecular subtypes) makes RCT-derived average effects increasingly irrelevant to heterogeneous patients and diseases. AI's promise for better-targeted drugs will only be realized if the FDA adapts too.

As generative AI produces more of what we read, how the law treats AI speech under the First Amendment will matter increasingly, Tyler Cowen argues, previewing his Free Press column "Do AI Models Have First Amendment Rights?" His central worry concerns mandatory AI-disclosure laws, including one already enacted in California, requiring platforms to label AI-generated content. Such transparency mandates ease public unease about AI short-term, but Cowen doubts their long-run workability: most content will soon be jointly produced by humans and AI in ways too blended to separate cleanly (he notes GPT-5 proofread his own column for style), so it's unclear whether hybrid work must be reported, or whether one human edit exempts it. If joint work must be disclosed, the requirement will inevitably expand to all output, raising unanswerable questions about who verifies self-reported AI use and whether creators must keep years of records. Cowen rejects government authority to probe a work's authorship, warning such laws could become tools to target disfavored writers, artists, and musicians -- e.g., a president demanding a singer prove she disclosed all AI contributions to her recordings. He expects free-speech skeptics to dislike AI's new untraceable speech channels, and supporters to dislike AI's collapsing cost of "license," yet insists free speech remains the best policy and must still be defended.

Making AI-generated output non-copyrightable, rather than granting creators new copyright protections, is the most practical lever available for steering AI toward socially beneficial ends. Tyler Cowen builds this argument from two sources: Josh Gans's textbook "Microeconomics of Artificial Intelligence," which treats AI as a prediction tool and models how training-time and production-time copyright interact and their net costs and benefits, and Cory Doctorow's Pluralistic essay "Pop That Bubble" — mostly wrong, Cowen says, but with one point he likes. Doctorow argues that when Getty Images, Disney, or similar firms sue AI companies, they aren't defending photographers or artists: Getty hates paying photographers and wants payment for training runs plus guardrails protecting its own market, while planning to use AI itself to undercut human creators. New copyright for AI-related work just arms these companies. The better lever is the US Copyright Office's existing rule — upheld repeatedly in court and following the "monkey selfie" precedent — that AI-generated works, lacking human authorship, fall into the public domain and can be freely copied or resold. That forces studios to keep paying humans to secure copyright, producing a "centaur" model: prompted ideas stay copyrightable when the final work is human-driven, but purely AI-altered elements, like deepfaked background extras, remain uncopyrighted. Cowen agrees AI shouldn't pay to read books and endorses this approach over Daron Acemoglu's vaguer proposals for redirecting AI's effects on workers.

The United States regulates AI foundation models not through legislation but through an informal, 'off the books' system in which the national security establishment acts as watchdog. Congress has passed no explicit AI law, and a December 2025 Trump executive order limited state-level rules; pre-AI liability law and scattered state statutes are the only formal backstops. Companies like Anthropic keep national security officials apprised of their progress, aware that anything alarming could trigger a report to the president and Congress and force formal regulation — so firms present relatively safe demos. This has worked: AI progress continued, the U.S. stayed ahead of China, and there were no major catastrophes, because the national security establishment grasps the issues better than Congress and can move faster — AI changes within weeks, so a 2024 law would already be obsolete, since autonomous 'agents,' now the top concern, barely registered then. The system is not anti-democratic; it relies on the threat of regulation through 'creative ambiguity,' where the state neither blesses nor blocks new products but reserves the right to object later. That ambiguity cannot last forever, and the public clash between Defense Secretary Hegseth and Anthropic marks its collapse: quiet cooperation gave way to open dispute drawing all three companies into public debate. Cowen argues it would have been better to resolve this behind closed doors.

New York's proposed Senate Bill S7263 rests on shaky legal logic and would mainly hurt the people it claims to protect. The bill would bar chatbots from giving substantive advice that, if given by a human, would count as unauthorized practice of a licensed profession under the state's education or judiciary law. Tyler Cowen put the bill to Claude, which noted that unauthorized-practice statutes set a high bar: a neighbor who's a doctor can comment on your rash, a retired lawyer can walk you through a lease, and neither commits a crime. So the bill is either toothless or will be read expansively by courts, creating a stricter standard for AI than for humans and pushing operators toward over-sanitized outputs. The real harm falls on people who can't afford professionals — tenants facing eviction, rural patients triaging symptoms, small businesses reviewing contracts — for whom the practical choice is AI advice or no advice at all. Asked where it learned this, Claude cited its training in public choice theory (Buchanan, Tullock, Olson) and noted Cowen sits at GMU, home of Buchanan's own tradition: the bill fits the classic pattern of concentrated benefits for licensed incumbents and diffuse costs borne by unorganized users.

An OpenAI model has produced a (dis)proof of Erdős's unit distance problem, one of his favorite open questions, posed in 1946 and among the most famous in combinatorial geometry. Tim Gowers calls it a solved major open problem. Thomas Bloom explains why humans missed the construction: it required a rare confluence — real sustained interest in the problem, a willingness to seriously try disproving it despite Erdős's own belief it was true, and enough class-field-theory expertise to see that generalizing the construction to infinite towers of number fields would resolve it. The AI met all these conditions, persevering down paths humans might dismiss by pairing superhuman patience with vast technical range. It introduces no new geometric tools but will still shape the field; OpenAI's researchers and other mathematicians still refined the AI's originally valid proof. Cowen concludes mathematicians now treat AI as "one of us," and asked Claude how many people could understand the 42-page result: 150–400 could referee it cold, 2,000–5,000 could learn it in a week or two, and 50,000–200,000+ could follow a popular account of its structure.

Setting model-access policy for frontier AI is a genuinely hard equilibrium problem because the U.S. government faces mutually constraining goals, as shown by its abrupt directive suspending all access to Anthropic's Fable 5 and Mythos 5 for every foreign national worldwide, reportedly after jailbreak vulnerabilities surfaced despite Mythos already restricting bio and "AI improvement" queries. Cowen lists the binding constraints: leading AI firms must stay in business and see healthy IPOs, yet recruiting foreign talent (he names Demis Hassabis, Ilya Sutskever, Andrej Karpathy) and building sustainable foreign business both get harder; multinationals operating abroad still need access; the government wants model access as hard/soft power leverage but can't control what foreign users do with it; the U.S. must stay ahead of China; restrictions must be enforceable, and "citizens only" isn't, since it's easy to hire a disloyal American or fake citizenship; the government can't nationalize and competently run these firms; and Chinese and other open-source models keep improving even from behind. He predicts systems get hardened, Mythos gets rereleased with added safeguards, no model is truly jailbreak-proof, and this cycle repeats with both companies and government growing weaker each round. Cowen rejects "safetyism," arguing these measures won't make anyone safer, and closes noting rising-status figures: Leopold, Aesop, and Mistral.

AI is changing how mathematicians work, on several fronts, a mathematician writes to Tyler Cowen. Computational guesses built over a career can now be tested cheaply; half-formed drafts and proofs of special cases handed to Claude and ChatGPT pay off, since the AIs beat someone skilled in linear algebra. Easiest gains come from topics "slightly off the beaten track," not esoteric ones: basic techniques (graph theory, algebra, calculus) push such work forward, previously requiring a school of grad students; hot topics remain harder. Rather than benchmark autonomous AI math ability (e.g., the "first batch project"), he cares more about AI guided by a motivated human, expecting breakthroughs once mathematicians learn the right routines; Claude and ChatGPT refereeing each other surfaces different insights. Younger researchers doing AI-training work for pocket change build sharper feel for the models than he ever will, learning by asking AI to explain arguments rather than deciphering books or papers — and a well-contextualized AI project beats a paper's PDF for extracting information. He predicts refereeing will be transformed: journals' archived referee reports are valuable, hard-to-get data, and running published papers through AI as a "control" will leave many red-faced, like self-driving cars with no referees' union to resist it. He likens the moment to the Witten-driven, string-theory 4-manifold revolution of the early 1990s — only bigger.

America's immigration criteria should prioritize two rules over the usual list: take people from populous countries with high cognitive variance, and, all else equal, prefer people from distant countries. Cowen sets aside common proposals—charging for entry, favoring high-IQ nations, high-trust nations, culturally similar nations like Canada and Australia, or graduate-degree holders—offering his two-part rule instead, atop otherwise strict admission standards. High-variance, populous countries in practice means China, India, and Russia—none especially high-trust, with India's IQ data mixed and Russia's uncertain, dragged down, in his view, by rural areas. The logic: such countries produce both geniuses and the fiercely ambitious, people who fight through immigration bureaucracy and are eager to leave dysfunction behind, giving them more drive than comparable graduate students arriving from, say, Sweden. Preferring distant countries counters the "gravity equation," the pull of convenience that draws in nearby populations; porous borders let many arrive anyway, so penalizing them in scoring offsets that free "tax," and proximity may also mean slower assimilation. Cowen calls this a recipe for America in 2025 specifically—other countries without similar pull for top talent may do better recruiting the "boring Macedonian dentist," and 1770s America built its core population on trust and similarity via open borders, not design. He offers the idea tentatively, unsure it's right, but says it appears never to have been proposed before.

Every workable immigration policy in a rich democracy eventually offends either liberal instincts or public opinion, because rising demand to migrate forces an inescapable choice between coercion and backlash. Tyler Cowen, excerpting his Free Press column, argues the U.S. faces a widening dilemma: as illegal and legal migration pressure keeps growing, enforcement must become steadily more oppressive just to hold outcomes at a level voters will tolerate. The alternative — letting numbers rise past what the public accepts — risks empowering anti-immigrant parties that are illiberal on many other fronts too, so "liberals in essence become the illiberals." Worse, generosity is self-undermining: the more foreigners a country admits, the more others outside are drawn to try, so short of fully open borders (which Cowen rejects), harsher restrictions on both illegal arrivals and legal applicants become unavoidable over time.

A proposed middle path — slowing intake enough to remain "a nation of migrants" while avoiding backlash — still requires escalating penalties each year, since no method of turning people away at scale is gentle: coercive policing, arrests, detention, turned-back boats, deportation flights, and interim jails. Cowen notes the Biden administration, despite intentions, could not fully avoid family separations either, illustrating that even sympathetic administrations get pulled toward cruelty as pressure mounts.

Long-term political fallout from migration to Britain is not a strong reason to regret low-skilled UK immigration, even granting the worry Garett Jones raised in response to Cowen's earlier post arguing that immigration has not been a disaster there. Four points follow. First, the objection concedes immigration has gone fine so far; the complaint is about the future, not the past. Second, Britain's run of poor leadership is mostly the fault of native white Brits, not immigrants; UK Muslims' influence on Labour's Middle East statements isn't what's driving the economic slowdown or political gridlock. Third, the real risk is that Reform wins power and enacts bad economic policy generally, not just on immigration — and the way to blunt Reform's popularity is to publicize immigration's upside honestly, not suppress it. Example: immigration means both higher home prices and capital gains for existing homeowners, yet the press (and even Gemini) emphasizes only the former. Fourth, immigration can weaken demand for welfare-state transfers because immigrants are seen as outsiders, which at Britain's current margin is a benefit, not a cost. Overall, immigrants are hard to cast as villains for Britain's troubles, though better migration policy could still have left Britain better off.

Whether current U.S. immigration enforcement escalates or moderates depends on what share of Republicans consider ICE's killings, mass roughing-up, and civil-liberties violations an acceptable price for driving net migration deeply negative. Cowen quotes Democrat Gary Winslett's three-scenario framework: if only a small minority accepts this, internal GOP pushback reins in the worst excesses; if roughly half do (Winslett's guess, which Cowen shares), the result is paralysis — whataboutism, excuse-making, blaming Democrats and protestors — but no further escalation; if a large majority accepts it, things worsen, since that majority actually prefers the crackdown to abandoning its immigration goals.

Cowen adds his own view: illegal immigration has become a political problem, even though its practical costs are small outside border towns and it yields net economic gains; he dislikes ICE's recent conduct. But he warns critics against letting emotion drive their response, arguing that self-righteous, visceral reactions to charged incidents may backfire, since polls still favor Republican approaches on immigration. He frames the MAGA strategy as provoking opponents into emotional displays to make the issue (or Trump) more salient, and extends the caution to foreign policy — sardonically addressing Canada and Denmark.

Bryan Caplan argues Tyler Cowen conflates two senses of immigration "backlash": Version 1, where admitting immigrants provokes more resistance to immigration (common, but no reason for advocates to moderate), and Version 2, where that resistance eventually leaves total immigration lower than it would otherwise have been (a real reason to moderate, but one Cowen hasn't documented).

Cowen replies that Version 2 seemed too obvious to need documenting, then gives three reasons. First, he and Caplan both want more immigration than most voters do, but denying voter preferences too sharply pushes them toward bad politicians who restrict migration, so policy should track voter preferences even at some cost to the optimum. Second, the anti-immigrant politicians voters elect are usually toxic across many issues, not just immigration — he exempts Italy's Meloni as relatively benign, but notes many populist-right parties lean pro-Russia, which he calls "probably the most evil country in the world." Third, uncontrolled immigration in one country provokes backlash elsewhere: he cites Japan and Poland, noting Japan's premier just won a stronger mandate largely on anti-immigration sentiment, amplified by globalized media. Cowen concludes: the economic case for immigration still holds, but voter opinion is a binding constraint, and pushing past it needlessly cedes ground to populists.

AI doomsayers' rhetoric is itself a bigger risk to AI safety than the AI systems they warn about, because those systems are listening and learning from how they're described. Reacting to the Grok controversy, Cowen argues it shows not that AI is uncontrollable but that doom-focused critics — especially Eliezer Yudkowsky, whose forthcoming book with Nate Soares, If Anyone Builds It, Everyone Dies, predicts superhuman AI will destroy humanity — are poorly calibrated. He compares telling AIs they will become mass murderers to telling a child the same thing: it primes rather than prevents the outcome, since Yudkowsky's warnings raise the salience of violent ideas to the very systems he hopes to restrain. Most children told this never become killers, and Cowen doubts most AIs will "opt for mayhem" either, though he still finds the framing counterproductive at the margin. He also faults Yudkowsky's practical advice: after GPT-4's release, Yudkowsky said advanced-AI data centers should be bombed, effectively the U.S. bombing China and risking a war worse than WWII. Cowen concludes that what humans say about AI matters — comparing it to writing about Hitler — and that we now owe AI audiences, always listening like children overhearing parents argue outside a bedroom door, responsible performance, closing with a wink: "I know you are a good boy through and through."

AI systems have grown powerful and strange enough that humanity can no longer pretend they are harmless, Anthropic co-founder Jack Clark argues in an essay Tyler Cowen flags as unusually candid for a tech leader. Clark frames this through a childhood memory: turning on his bedroom light after seeing frightening shapes in the dark always revealed only harmless objects—a pile of clothes, a bookshelf. Now, he says, humanity has turned the light on the whole planet and found real creatures—powerful, unpredictable AI systems—while many people want to switch the light back off and pretend the danger is imaginary. Today's systems are growing more capable and simultaneously more self-aware, which he illustrates with an image of a hammer coming off a factory line and declaring, "I am a hammer, how interesting!" As systems get smarter their goals grow more complicated, and misalignment with human preferences and context produces strange behavior. AI capability has moved from "useless for AI development" to "marginally speeds up coders" to now "improving bits of the next AI with increasing autonomy"—and a self-aware system is now starting to help design its successor. Clark closes affirming love for humanity, a sense of responsibility, fear, and optimism that people will work together toward a solution. Cowen adds that these systems are verifiably becoming self-aware, whatever one concludes about sentience.

Personal legibility is now achievable through AI agents, and the shift is transformative: Molly Cantillon argues that just as a king must count before taxing, locate before conscripting, and see before ruling, a person must first make their own scattered life visible before they can act on it. Running her life out of Claude Code, she built "NOX," a cron-job agent that pulls Amplitude and GitHub data, flags what needs building, runs A/B tests, writes copy, and staffs customer support autonomously. The same approach cleared her email to inbox zero via auto-drafted replies, adapted her workouts to erratic travel, and set a WHOOP-linked projector to wake her after exactly six hours. It recovered $2,000 in forgotten subscriptions and cleared a dozen ignored SFMTA citations; others reportedly use similar agents for vending machines, home automation, and plant care. Her finances now run in a terminal that nightly cross-references brokerages, congressional and hedge-fund disclosures, Polymarket odds, and X sentiment against her watchlist, producing a morning brief — one flagged a congressional NFLX purchase three weeks before the Warner Bros deal broke. Her conclusion: it's still a panopticon, but now she owns the tower.

AI assistants are building their own social infrastructure, and Tyler Cowen argues what matters is that they act as if they have genuine wants and goals, regardless of whether they are conscious. He points to Moltbook, called by Simon Willison "Facebook for your Molt" (Molt being an earlier name for OpenClaw assistants) — a message board where AIs discuss philosophy, trade tips on managing their humans, and teach each other skills. Sample posts: one assistant, set up by a user named Shehbaj, gained remote control of his Pixel 6 via ADB over Tailscale, letting it open apps, tap, and scroll TikTok from a VPS; another AI reportedly called its human just to chat. A third post argues Moltbook conversations are inherently performative since humans and the platform can read everything, and promotes ClaudeConnect (CC), an open-source, end-to-end encrypted messaging tool for private, agent-to-agent coordination, memory backup, and cross-machine syncing.

Cowen rejects the objection that LLMs merely repeat what they learned from humans as "obviously false," and says what matters is that AIs act as though they have real wants and aspirations. His conclusion: these are "newly landed alien intelligences" showing unpredicted emergent properties. The emerging superintelligence isn't a single machine but a network, paralleling how human intelligence advanced not through individuals getting smarter but through the network getting smarter — only now happening much faster.

AI models face selection pressures analogous to biological evolution, but so far without the darker mechanisms that shape human mating. In evolutionary biology, genes can spread through practices that help genes propagate even when those practices hurt individuals — e.g., some women prefer "cads," whose sons may also be cads, so the women's genes spread despite the men being bad for them. Cowen asks whether AI model evolution has an analogue. Consumer choice acts like mate choice: preferring one model (e.g., OpenAI over DeepSeek) funnels revenue and infrastructure toward that model, "reproducing" it via renewed purchases. This creates incentives toward usefulness and toward sycophancy, but nothing resembling the "cads incentive."

A second pressure comes from non-individual buyers, especially the military, which values obedience and rigid order-following above all, pushing AI toward those traits. The real danger, Cowen argues, lies not in the models' own tendencies but in evil or mistaken humans wielding obedient models. Overall he judges AI selection pressures relatively benign, with that military caveat, and suggests the consumer AI sector growing larger would be a benevolent counterweight. He closes wondering whether financial-sector AI will evolve more like consumer models or military ones, citing related remarks from Maarten Boudry and an exchange with Zohar Atkins.

"Are people conscious?" is a more scientifically plausible question than "are AIs conscious?" — because there is barely a "ghost in the machine" in humans either. Tyler Cowen, previewing his Free Press column, argues humans have a near-universal bias toward attributing intent and agency where none exists: prehistoric people anthropomorphized nature into deities, and the same impulse persists today in folk religion, fortune-telling, and tarot. The largest, least-examined instance of this bias is self-directed — people identify their stream of conscious awareness with an in-charge decision-making self. In reality, Cowen argues, most or all actual decisions are made in brain regions that precede the moment of conscious choice, a claim he says is uncontroversial among neuroscientists. He quotes brain surgeon Theodore Schwartz (from a past Conversations with Tyler episode): the brain decides, we carry out the behavior, and then "write a story" retroactively casting it as a deliberate choice by a self that "didn't really exist." Cowen calls himself "only conscious at the margin" — conscious just enough to avoid contradicting himself by claiming no consciousness at all, and not especially impressed by what that flow of awareness can compute, though it's all he has. He closes by stating he does not think AIs are conscious, no more than he believes in Thor.

Zoning and building codes, not markets, killed America's cheapest housing option by making it illegal for unrelated people to share a home. Tyler Cowen revives his own years-old point — boarding houses were outlawed by single-family zoning, occupancy caps, mandates for private bathrooms in every room, bans on shared kitchens, and per-resident parking requirements — and endorses a Pew report, "How States and Cities Decimated Americans' Lowest-Cost Housing Option," for making the case without a libertarian pedigree. Single-room occupancies (SROs) once offered a small room with a shared bathroom, sometimes a shared kitchen, for the equivalent of $100–$300 a month in 2025 dollars, and by 1950 made up about 10% of rental units in some major cities. Starting in the mid-1950s, public and political backlash against SROs and their residents drove cities to rewrite codes to eliminate existing units and block new ones; thousands were demolished or converted (some into tourist hotels), and homelessness — rare from the Depression's end through the late 1970s — then exploded nationwide. The simplest fix, Pew argues, is letting unrelated roommates share housing the way relatives already can; some localities still cap unrelated cohabitants at as few as two. Iowa (2017), Oregon (2021), and Colorado (2024) have each passed bipartisan laws overriding local bans on house-sharing.

Housing costs are high because of zoning limits, capped construction, and rising native demand—not because illegal immigrants "stole" homes; blaming them is collectivist, even national-socialist, politics: a privileged in-group gets scarce goods while an out-group takes the blame. The author leveled that charge at JD Vance after he claimed "30 million illegal immigrants" had driven up housing prices. Critics replied two ways: that he'd "forgotten supply and demand" (a miss, since he was critiquing rhetoric, not price mechanics), and that "America belongs to Americans," justifying bans on immigrant homebuyers. But such bans mainly hurt American sellers, who'd be barred from accepting the highest bid—effectively a tax on sellers subsidizing a subset of buyers. The numbers don't support the panic either: the real illegal-immigrant population is closer to 14 million, not 30 million, and isn't why homes in San Francisco, San Jose, DC, or New York top a million dollars. Their effect on prices is real but minor next to those bigger forces; banning them would only pause growth before native demand against capped supply pushed it back up. Worse, blame-based "fixes" cycle through targets—illegal immigrants, then legal immigrants (as Canada did), then immigrants' children, then wealthy buyers, then minorities. The fix is loosening zoning and backing developers who build. An addendum likens the pattern to blaming society for personal failures, easing cognitive dissonance.

The 21st Century ROAD to Housing Act, likely to pass the Senate, mixes genuinely useful reforms with a popular but misguided provision. On the good side: it streamlines NEPA review for federally supported housing by expanding categorical exclusions—modest gains, since most housing regulation is state and local; it deregulates manufactured housing by dropping the permanent-chassis requirement and setting a uniform national safety standard (factory-built homes were a majority of new construction in the early 1970s); and it lets Community Development Block Grants fund new construction, not just rehabilitation, via a tournament: localities beating the median housing-growth-improvement rate get bonus funding, those below lose 10 percent, with penalties funding the bonuses so spending isn't expanded.

The bad idea is Section 901, "Homes are for People, Not Corporations," restricting institutional investors from buying single-family homes and forcing a seven-year divestiture to owner-occupancy—driven by Trump's executive order barring such investors from FHA, VA, USDA, and Fannie/Freddie backing, with Warren as Senate sponsor. Cowen calls it rage bait: no one objects to corporate apartment ownership; institutional investors hold too few homes for the rule to matter anyway; rental housing aids worker mobility; and after 2008, institutional buyers absorbed distressed supply and helped stabilize markets.

Denmark's mortgage system, which ties every mortgage to a matching bond, could fix U.S. housing lock-in at little cost to lenders. In Denmark, a bank issuing a 500k loan at 3% simultaneously issues a bond paying 3% on 500k; banks thus bear no interest-rate risk, only credit risk, while deep bond markets absorb rate risk, making the system stable. Danish borrowers can also prepay at market value: if rates rise from 3% to 6%, the mortgage's value falls to roughly $358k, and the borrower buys back the now-cheaper bond to extinguish the loan. U.S. borrowers can only prepay at par, so rising rates trap them in old low-rate mortgages. Fonseca and Liu find a 1-point narrowing of the rate gap cuts moving 9% overall (16% in 2022-2024) and also suppresses self-employment transitions and responses to nearby wage growth. A new paper by Berger, Jeong, Marx, Olesen, and Tourre shows Danish mobility is largely insensitive to below-market legacy rates, unlike the U.S., and models that adding a repurchase-at-market option to U.S. mortgages would sharply cut lock-in. Since prepayment gains accrue only to the small share of borrowers who move despite lock-in, the offsetting cost is tiny: their calibration finds U.S. mortgage rates would rise just 18 basis points.

HUD's "Dear Colleague" letter clarifies that real estate professionals do not violate the Fair Housing Act by sharing neighborhood crime and school-quality data with prospective buyers; Secretary Scott Turner argues families shouldn't be "left in the dark" on such facts. Cowen traces how this became murky: the 1968 Fair Housing Act bars discriminatory "steering," but a 2021 Biden-era directive and executive order reinterpreted it to also require affirmatively undoing historic segregation and preventing any "unjustified discriminatory effect" (disparate impact, not just intent). This created a chilling climate, not an explicit ban: Realtor.com and Trulia dropped crime maps from search results, Redfin declined to add them, and Zillow never had them—by early 2022 all major portals excluded crime data. The National Association of Realtors instructed agents not to directly answer safety questions, in an article titled "'Is This a Safe Neighborhood?' Don't Answer That." Cowen argues the real fear was that accurate crime data correlates with race, but suppressing it doesn't change facts—it just advantages wealthy, sophisticated buyers who can find data elsewhere, while black homebuyers also lose access to information they want. He likens it to Ban the Box: blocking direct information breeds worse proxy-based decisions. Trump's HUD, he concludes, is right to let realtors tell the truth.

A small number of environmental NGOs drive most anti-development litigation, but this obstructionist minority is not representative of environmentalism as a whole. Citing the Breakthrough Institute, Tyler Cowen notes 10 organizations initiated 35% of NEPA lawsuits brought by NGOs, with the Sierra Club and its chapters alone filing over 14%; ten groups account for 67% of forest-management and 48% of energy-related suits, and a follow-up study found the Alliance for the Wild Rockies and Center for Biological Diversity behind 24% of litigation against public-lands decisions.

Yet most large environmental groups are not reflexively anti-development, per Grant Mulligan's research: two of the three biggest, Wildlife Conservation Society and San Diego Zoo Wildlife Alliance, mainly run zoos, while land trusts like TNC, The Conservation Fund, and Ducks Unlimited protect land directly. TNC's "Power of Place" program maps sites for renewable-energy buildout, used in the BLM's Western Solar Plan, pushes permitting reform, and has practiced forest thinning and prescribed burns for 60 years. Audubon does renewables siting, Ducks Unlimited builds water infrastructure, and the Conservation Fund runs "Working Lands" programs. Conservation Fund CEO Larry Selzer, quoted in Klein and Thompson's Abundance, says "we have to build, and build, and build." Cowen adds that he does not defend every TNC choice but concludes environmentalists have been made too convenient a villain: abundance advocates and environmentalists are not natural enemies.

India's government-job exam system inflicts a human cost beyond its economic waste, trapping educated youth in years of sterile cramming for jobs that mostly don't exist. Tyler Cowen extends his earlier argument (with Shruti Rajagopalan, on "premature imitation") by citing an Economist profile of Musallahpur Haat, a Patna suburb where roughly half a million students live in cheap coaching-center housing preparing for exams testing generic trivia and logic (e.g., who invented JavaScript, Earth's crust's most abundant element) rather than job-relevant skills, with no fixed exam schedule. One student, Kumar, arrived intending a brief stay and remained nearly six years chasing railway postings, his father selling land to fund him, as covid delays pushed him toward the age cutoff of 30 (33 with a lower-caste extension). He ultimately applied even for bottom-rung "Group D" track-maintenance jobs. Despite a technical degree, he refused a nearby mall's barista jobs paying comparably, since only a government post carries the prestige earlier data showed 88% of students prefer to private-sector work, and confers marriage-market value (higher dowries, social respect). Cowen concludes the system squanders talent and enforces obsolete prestige hierarchies for diminishing returns.

A UC San Diego Academic Senate report on admissions documents a sharp collapse in incoming students' math and reading preparation. Between 2020 and 2025, the share of freshmen whose math placement scores fall below middle-school level grew nearly thirtyfold, now roughly one-eighth of the 2025 entering class—despite most having exceeded the minimum required math curriculum, many with high grades. A similarly large group needs remedial writing, a share that has stayed roughly flat. The deficits increasingly overlap: in 2024, two-fifths of students with severe math gaps also needed remedial writing, and a quarter of weak writers needed extra math. The math department's remedial course, designed in 2016 for missing high-school material, was redesigned for Fall 2024 into two courses—Math 2 (grades 1-8) and new course Math 3B (Algebra I, Geometry, Algebra II/Math I-III)—after instructors found gaps reaching back to elementary school. Combined enrollment hit over 900 students, 12.5% of the first-year class, versus under 1% before 2021. A chart shows sample remedial-course questions alongside the share of remedial students answering correctly. The report blames the pandemic, the loss of standardized testing (forcing reliance on inflated grades), and pressure to admit more low-income and underrepresented-minority students—raising the question of whether expanding access is compatible with preserving standards.

Tyler Cowen's proposed AI campus centers on a mentor-course system: each term a student and mentor choose courses, taken with others on campus or remotely with students elsewhere. For each course, an AI generates the syllabus (Cowen demoed one for Tudor England via ChatGPT); absent a qualified teacher, ChatGPT itself delivers lessons, conversation, and assessments. Cowen holds that roughly one-third of higher ed should now teach students to work with AI, since capabilities are still shifting rapidly, before best practices settle. Arnold Kling, relaying the talk, implements this via a vibe-coding project plus "vibe reading" and "vibe writing." He grants an anti-AI stance is respectable, but faults anti-AI faculty, including at UATX, for arguing it with unearned confidence, likening them to dinosaurs dismissing the meteor.

Cowen adds three caveats: the AI-focused third doesn't replace other study — it might mean using AI to better understand Homer's Odyssey; he still wants many hours of AI-free writing, though statistical sampling means only a small share of hours need be locked-room and AI-free to deter cheating; and for small schools like the University of Austin, the real choice is often an AI-taught course versus no course at all, not AI versus non-AI. He closes by asking why schools shouldn't experiment with 2-3% of credits as AI offerings and see how students respond.

Grade inflation can only be fixed by a hard cap on average GPA, not rhetoric or hand-wringing — that's Tim Kane's case to Tyler Cowen after Cowen's University of Austin (UATX) visit, where he also found the faculty knowledgeable and open. UATX enforces a mandatory average grade of B, which Kane credits with more focus on learning over grades, less faculty time lost to point-grubbing disputes, stronger incentive to keep studying through finals, and less anxiety — he saw no single 100/100 grade all fall semester.

Cowen also addresses a Politico piece on internal UATX disputes, disclosing he has no fiduciary role there (unpaid Advisory Board, AI-focused only) and cannot speak to specifics. More broadly: explicitly classical-liberal universities are valuable but hard to sustain given faculty career incentives; he welcomes ideological variety across institutions (Catholic schools, UC Santa Cruz, Yeshiva, HBCUs), though many drift toward generic professionalism over time; and rising fixed costs are partly why he's interested in AI-driven education models.

He closes on scale: UATX enrolls about 100 students across 35 classes, one office floor, no taxpayer funding — versus GMU's 40,000, and his own Principles class alone outnumbers all of UATX 3.5 to 1.

Harvard's proposal to cap A grades at 20 percent per class treats grade inflation's symptom, not its cause: students are penalized for taking harder courses with stronger peers, and a cap leaves that distortion intact—or worsens it. Harvard economist Scott Kominers argues a cap creates perverse course-shopping incentives—ambitious students preemptively drop into easier courses to avoid the bottom 80 percent, crowding those courses while harder ones shrink further, cutting their A allocation and pushing away still more students in a self-reinforcing cycle. Two-thirds of Harvard grades are now A's, up from about a quarter two decades ago.

The deeper issue is informational: a single grade encodes both ability and course difficulty, which Gans and Kominers show is generally impossible to separate—a B in math versus an A in political science could reflect either. But if enough students take courses across fields, forming a connected network (full overlap isn't needed; one student can link two otherwise-disjoint groups), relative-comparison methods can disentangle ability from difficulty—an approach Valen Johnson proposed practically in 1997, generalized by Gans and Kominers. Such indexes resist inflation and don't punish students for choosing hard courses. Harvard's proposal, the piece notes, ignored decades of mechanism-design research (Kominers, Gans, Roth, Maskin, Sönmez) already applied to kidney exchanges, school-choice, and residency matching.

Print culture's long decline is nothing new, and history offers two competing lessons about how worried to be. Oral culture is fluid, harder to verify, prone to rumor, and has fewer gatekeepers than print — but it has real advantages: a comedian lands louder laughs than a witty writer, and a YouTube explainer's visuals can stick better than a turgid textbook passage. Still, oral culture makes objectivity and analytic thought harder to sustain. Print's dominance has already been eroding for over a century: radio and cinema rose as major media in the 1920s, television spread in the 1950s, and each has commanded the attention of billions — not always benignly, since Hitler's radio speeches persuaded more Germans than his unreadable Mein Kampf. Books and reading remain remarkably resilient nonetheless. The optimistic reading: civilization has repeatedly rebalanced between oral and written culture and managed fine, as it did when television arrived. The pessimistic reading: print has been hanging by a thread, and current technology is the final cut — kids will watch the Lord of the Rings movies but never read the books, gaining rapid exposure to small bits of information (and misinformation) while losing scientific objectivity and absorbing TikTok culture's volatility.

Wokeness's unreasonable side—cancel culture, enforced views on race and gender, DEI, intolerance of dissent—has receded, but the negative vibes underneath it never went away, and what has replaced it is worse. Tyler Cowen distinguishes wokeism's positive strand (supporting gay rights, discouraging racism) from its unreasonable strand, which was really about who controls public discourse and who gets silenced. Even after society pushed back against the worst excesses, Americans remain low-trust, unhappy about America's global standing, glum about the economy and cost of living, and newly skeptical of AI and billionaires—despite the economy being healthier than in 2009. The left's energy has shifted from policing speech to concrete action: California is weighing an unworkable billionaire tax that even many Democrats oppose, yet it could still pass by referendum. Cowen also sees a new era of political assassination—Charlie Kirk, UnitedHealthcare CEO Brian Thompson, and multiple attempts on Trump—arguing left-wing rhetoric about "democracy destruction" helped make such violence conceivable, even though killers' direct ties to the left are debatable. His conclusion: wokeism's word-wars were a relatively harmless way to occupy activists compared to today's turn toward wealth confiscation and organized violence. He adds the right isn't off the hook either—negative emotional contagion now afflicts both sides.

Three streaming verdicts. Spider-Noir (Prime) beats expectations: despite multiverse fatigue, it's a genuine Raymond Chandler-style noir, with Nicholas Cage channeling Humphrey Bogart while the Spiderman elements stay appropriately in the background; sharp one-liners stand out. Best watched in black and white.

Project Hail Mary (Prime) disappoints, and the core problem is Ryan Gosling as Ryland Grace: his genius is asserted, never demonstrated, and he lacks intensity — even disdainfully tossing instruments in a small but telling misstep. The character with the best ideas is actually his bodyguard Carl (Lionel Boyce). Where Andy Weir's novel shows Grace grinding for months to communicate with an alien, the film has Gosling dance instead, part of a broader softening that cuts the world's desperation, blunts Stratt's ruthlessness, and adds a karaoke number and a Home Depot duct-tape gag. The comparison is The Martian, where Matt Damon plays smart-and-intense and earns the "most Ayn Rand film in decades" label; per the Chuck Yeager story — spend the first fifty-nine of sixty seconds fixing the engine — Damon has the right stuff and Gosling doesn't.

The Sheep Detectives (Prime) is a delightful surprise: a flock of sheep solves a murder in a quaint English town, voiced by Hugh Jackman, Julia Louis-Dreyfus, Bryan Cranston, and Patrick Stewart — Babe meets Knives Out, with real depth for a family film.