Notes — Nate Silver on Life’s Mixed Strategies

Notes on Nate Silver in conversation with Tyler Cowen — Conversations with Tyler (https://conversationswithtyler.com/episodes/nate-silver-3/), 13 August 2025.

Four questions [Adler frame]

Q1 — What is it about as a whole? A wide-ranging conversation built to mark the paperback of Silver’s On the Edge: The Art of Risking Everything, using poker and gambling as the running lens on how people should reason about risk, probability, and prediction. It moves from game theory at the poker table to expected-value reasoning about longevity and Pascal’s wager, to the mechanics of elections and prediction markets, to a sceptical reading of AI timelines, and finally to sport, the Bay-Area rationalist subculture, and the politics of immigration and populism. The through-line is Silver’s poker-player epistemology: local, specific, actionable information beats grand theory.

Q2 — How is it argued? By worked example and calibrated numbers rather than doctrine. Silver reasons from the mechanics of a specific domain — the mixed strategies a poker solver reveals, the 55/45 edge that makes a winning gambler, the 90 per cent accuracy you get on a voter from just name and zip code — outward to a general claim. He states probabilities explicitly (P(Pascal) = 0.001; Democrats 55–45 for 2028; AI at ‘the 40th percentile’ of expected progress) and he prizes the ability to randomise deliberately. Cowen presses with binary or contrarian framings; Silver characteristically answers with a distribution and a hedge.

Q3 — Is it true, in whole or part? Strongest where Silver is inside his craft — poker game theory, sports betting edges, election polling and its calibration baseline back to 1936 — and appropriately tentative where he flags he is speculating (thermostatic politics, immigration reversing ‘within a few years’, which NBA team LeBron should join). His central methodological claim — that a Substack post or a book now does the work a journal article used to, and that ‘90 per cent of academic papers would work perfectly fine as blog posts’ — is a provocation stated as such, not a proven finding. [?] His AI-timeline scepticism (no near-term intelligence explosion; superforecasting AI still 10–15 years out) is a bet against Cowen’s one-to-two-year view, and is argued rather than settled.

Q4 — What of it? A transferable model of decision-making under uncertainty: treat life as a series of mixed strategies, weight local and actionable information above elegant theory, and hold your beliefs as calibrated probabilities you would bet on. It also gives the wiki a working sceptic on the intelligence-explosion thesis, a counterweight to the Deciding Under Uncertainty and AGI clusters, and a natural neighbour to the decision-theory material around Kill Criteria, Risk Posture, and Effective Altruism.

Glossary

Mixed strategy — a decision rule that deliberately randomises between options so an opponent cannot predict you; game theory shows the optimal poker play is very often a mix rather than a single ‘right’ move, so a player may literally use the tournament clock as a random number. [§ On game theory, poker, and reading tells]

Nash equilibrium — the stable state of a game where every player is playing optimally given everyone else’s choices, so no one can gain by changing alone; solving it for poker revealed how many hands must be played as mixes. [§ On game theory, poker, and reading tells]

Expected value (EV) — the probability-weighted average payoff of a choice; Silver’s habit of mind is to act on the option with the highest EV, though he stresses hyper-optimising every parameter of daily life tends to make you miserable. [§ On expected value, Pascal’s wager, and the shape of risk]

Pascal’s wager — the argument that one should believe in God because the possible infinite upside outweighs the finite cost; Silver puts the probability at 0.001 and declines to act on it, preferring the ‘terrestrial tractable realm’. [§ On expected value, Pascal’s wager, and the shape of risk]

Condorcet winner — the candidate who would beat every other in a head-to-head vote; Silver’s objection to ranked-choice voting is that it is not Condorcet-optimal, since the order of elimination can knock out a compromise candidate (his example: Brad Lander) who would have won any pairwise contest. [§ On elections, voting systems, and prediction markets]

Superforecaster — a person with a demonstrated track record of unusually accurate probabilistic predictions; Silver argues AI will not match human superforecasters for 10–15 years because forecasting demands broad general knowledge and rapid adaptation to changing inputs, not static problem-solving. [§ On AI progress, superforecasting, and the limits of AGI]

P(doom) — a person’s estimated probability that advanced AI leads to catastrophe; Silver reports his rose relative to the book’s first edition, driven less by an intelligence explosion than by humans using AI to make weapons or terrorism easier. [§ On the rationalist bubble and West Coast culture]

Thermostatic effect — the tendency of public opinion to swing against whatever the status quo is, like a thermostat correcting temperature; Silver invokes it to argue populist sentiment may have passed a high-water mark in high-income countries. [§ On populism, immigration, and thermostatic politics]

Key claims by section

On game theory, poker, and reading tells [§ On game theory, poker, and reading tells]

Solving the Nash equilibrium for poker showed that nearly every hand is a mix of some kind; against any opponent tell or read, a player moves from literal indifference toward a dominant strategy, so reads and table image matter a great deal.
Silver will randomise deliberately — using the tournament clock or the rotation of his chips — and can spot others doing the same.
Reading tells is highly contextual and semantic: a rapid heartbeat can signal a strong hand or a bluff, so a tell must be correlated with the specific player and situation. Being right 60 per cent of the time versus a 50 per cent baseline is a huge edge; in gambling any 55/45 edge is enormous.
Poker has gone furthest of any field in the literal real-world manifestation of game theory; you see weaker versions in football (a draw play on third-and-long has higher EV because it is unexpected).

On academic influence [§ On academic influence]

Silver reads Substacks over journal articles and thinks academics have lost influence, partly by migrating to Bluesky and partly by becoming reflexively, brain-meltingly anti-Trump — the ‘slow-cooking’ method of academia is a poor match for the news cycle.
The right unit is no longer the paper: a multi-day Substack post, a book that goes deeper than anyone before, or investigative journalism. He estimates 90 per cent of academic papers would work fine as blog posts, minus the Greek-symbol pretence. [?]
A newsletter’s value is tonal control — signalling when he is speculating, joking, or presenting an original finding — which the sterile academic register loses.

On expected value, Pascal’s wager, and the shape of risk [§ On expected value, Pascal’s wager, and the shape of risk]

Hyper-optimising every daily parameter (Bryan-Johnson-style longevity) probably makes you miserable; Silver makes ‘serious but incremental’ improvements instead.
He puts Pascal’s wager at p = 0.001 and declines to act on it, preferring tractable terrestrial problems over ‘infinite ethics’ edge cases like shrimp welfare (while conceding it is good that some people worry about them). He has shifted from atheism to ‘affirmatively agnostic’, partly because AI should make people ask more questions about the nature of reality.
Most people segregate risk rather than holding one integrated attitude (Silver’s example: Ezekiel Emanuel avoiding restaurants but riding a motorcycle); people are not very meta-rational about risk. Apparently irrational behaviour like loss aversion is often rational on a higher plane, serving an evolutionary or disciplining purpose.
Flying ‘too close to the sun’ after success is dangerous because feedback degrades and you surround yourself with yes-men (his read on Elon Musk); confident players play better, but sycophancy is corrosive.

On elections, voting systems, and prediction markets [§ On elections, voting systems, and prediction markets]

Silver has turned mildly anti-ranked-choice-voting: it is not Condorcet-optimal (elimination order can drop a compromise Condorcet winner like Brad Lander) and it is slow to count. He wants a US norm of counting votes within 24–48 hours, as India does, and better one-question exit polls of the European/Latin-American kind.
On proportional representation he leans yes as a critic of both parties, but notes America’s many veto points have probably been good for capitalism and growth.
Prediction markets (he consults for Polymarket) have improved and grown more liquid but still misprice — e.g. giving Zohran Mamdani, born in Uganda and thus constitutionally ineligible, a 7 per cent chance of the Democratic nomination, or overpricing a third Trump term.
With just a name and zip code he can predict a voter’s choice with ~90 per cent accuracy, but elections turn on 1–2 per cent margins, so blunt AI-on-Facebook-posts models would not beat polling, which has a calibration baseline back to 1936. He would give AOC his single free bet for the 2028 Democratic nomination (no one above 15 per cent).

On AI progress, superforecasting, and the limits of AGI [§ On AI progress, superforecasting, and the limits of AGI]

Silver holds to his prior estimate that AI matching human superforecasters is 10–15 years out; he rates the last year’s AI progress at the 40th percentile of what he expected. Cowen counters with one-to-two years and calls him too pessimistic.
His key distinction: static problems (Math Olympiad, chess) differ from dynamic systems with changing inputs. LLMs are very bad at poker because it needs rapidly developed exploitative strategies against an evolving game-theory dynamic; he worries Math Olympiad results reflect ‘teaching to the test’.
He separates rough AGI for desk jobs from superintelligence and from AGI for physical labour, and thinks people leap between them too quickly; he is now comfortable saying an LLM ‘thinks’ and that they do reason, while doubting an emergent intelligence explosion.
Training on a thought process is hard: he can train himself on his partner’s good eye for art but that caps him at a B+, and ‘a B+ trader is often on the losing side of a trade’.

On sport — the NBA, betting, and coming out [§ On sport — the NBA, betting, and coming out]

Since Jason Collins (2013) no NBA player has come out as gay despite gay marriage’s normalisation, whereas ~44 WNBA players are out; Silver offers selection effects (elite sport tracks athletes early; distractions to identity can knock you off a narrow path) and a recent conservative backlash in sports, calling the gap genuinely surprising.
Injury data for women’s sports is much worse than for men’s, so a diligent modeller combining a model with league knowledge can still find edges (women’s NCAA, WNBA) before markets adapt — though any consistently winning bettor gets limited quickly.
On the draft lottery Silver likes the Mike Zarren ‘wheel’ (each team picks first once every 30 years, known in advance) and dislikes incentivising failure; he notes American sports are ‘socialist’ and European sports ‘capitalist’. He would trim the NBA regular season toward ~72 games and cap individual load (e.g. 75 games, guaranteed rest published in advance) as continuous high-intensity play strains players more than other sports.
He argues the ‘only a few teams can win’ thesis is overstated (Nuggets, Raptors, Giannis as counter-examples) and that people over-index on the Bulls and Warriors dynasties.

On LeBron, mentors, and top talent [§ On LeBron, mentors, and top talent]

Silver reads LeBron James as the best NBA player of all time by career metrics but never able to win hearts short of three more titles; he would send him to a young team (the Spurs) to win and mentor, and thinks a Wemby-led Spurs could plausibly contend now — Cowen disagrees sharply.
On betting on flawed talent: Silver wants high variance, favouring a player with a fixable problem (Luka Dončić) over one already maximising, because the upside is larger; Cowen tends to bet against low-conscientiousness players.
His mentors are Bill James (technical subjects can still be well written; an 8/8 writer-statistician can beat a 10/2), Richard Thaler, and Cowen himself; on AI he treats Zvi Mowshowitz as a comprehensive, level-headed guide despite Mowshowitz’s high P(doom). He notes he has always ‘blazed my own trail’ because the product he wanted did not exist.

On the rationalist bubble and West Coast culture [§ On the rationalist bubble and West Coast culture]

Silver calls the AI-safety/rationalist community a bubble — proudly weird, experimental (polyamory), open-minded, but underweighting political constraints and human adaptability; they leap too fast from rough AGI to physical-realm AGI to ASI.
His P(doom) rose in the new preface (written Feb–March, during Musk’s White House influence and the tariff push), driven mainly by humans using AI for weapons, terrorism, or dangerous compounds rather than by an intelligence explosion; asymmetric technology where one person could do great harm (Bostrom) worries him.
EAs and rationalists are high in openness and high-variance in conscientiousness, prone to gullibility (he thinks EA should have taken more reputational damage over Sam Bankman-Fried); poker players share the phenotype but are more suspicious and street-smart.
The West Coast is diverging culturally from the rest of the US (a Bay-Area house party may lack wine; Seattle ‘feels like Canada’); doom-belief correlates with polyamory partly through hedonism and disconnectedness, though Cowen finds the economistic explanation too neat.

On populism, immigration, and thermostatic politics [§ On populism, immigration, and thermostatic politics]

Canada cannot win the Stanley Cup (none since 1992); Silver ties it partly to tax law — no state income tax in Florida within a hard salary cap effectively pays players 10–15 per cent more.
He mostly believes in thermostatic effects — opinion runs against the status quo — and suspects populist sentiment may have passed a high-water mark in Canada and Western Europe, while conceding Cowen’s point that it is still rising elsewhere. [?]
On the puzzle of why European centrists do not simply cut the immigration voters say they want: Denmark and Trudeau’s Canada have moved; the left was historically more restrictionist (Bernie Sanders) over welfare-state cost. He predicts the fertility crisis and ageing populations will reverse the politics ‘within a few years’ toward wanting skilled young workers. [?]
US politics is unusually efficient at producing consistent ~50–50 coalitions; America is currently quite pro-immigrant relative to the world. On polling, voters side with the left on itemised immigration questions yet trust Trump more overall, because they fear progressive excess (‘give them an inch and they’ll take a mile’), a worry Silver traces to COVID-era public-health overreach.

Nate Silver on Life's Mixed Strategies