Notes — Nate Silver on Risk-takers, Politicians, and Poker Players

Notes on Nate Silver in conversation with Tyler Cowen — Conversations with Tyler (https://conversationswithtyler.com/episodes/nate-silver-2/), 21 August 2024.

Four questions [Adler frame]

Q1 — What is it about as a whole? A conversation timed to the hardback launch of Silver’s On the Edge: The Art of Risking Everything, using the book’s cast — poker players, sports bettors, sharps, hedge funders, effective altruists — to probe what risk-taking temperament is, where it comes from, and what it is good for. It runs from counterfactual life-simulation and status-quo bias through the economics of sports betting, the game-theoretic professionalisation of poker, the failure of FiveThirtyEight to become a real business inside Disney, the homogenising effect of analytics on sport, Sam Bankman-Fried as a case study in risk-taking gone wrong, and closes on AI timelines for superforecasting and the contingency of political history.

Q2 — How is it argued? By anecdote, worked example, and Silver’s own self-report rather than formal argument. He reasons from his own biography (the 2006 poker crackdown that pushed him into founding FiveThirtyEight, his Enneagram scores, his father’s Sovietology) and from the book’s interview subjects (Peter Thiel, Sam Bankman-Fried, mountain climbers, an astronaut) to general claims about a ‘risk-taking gene’ paired with analytical skill. Cowen presses with pointed, sometimes adversarial framings — asking Silver to rate his own status-quo bias, defending a strict buy-and-hold libertarianism against Silver’s more permissive view of gambling, and pushing on whether ‘the river’ description of Silicon Valley, EA, and poker is too tidy a single category.

Q3 — Is it true, in whole or part? Strongest where Silver reports what he has directly lived or reported first-hand: the mechanics of sports-betting price discrimination against sharps, FiveThirtyEight’s incentive failure inside a company ‘printing money’ from theme parks and cable, and the SBF material, which draws on Silver’s own interviews with Bankman-Fried for the book. Weaker and more speculative where he reaches for a ‘thrill-seeking gene’ or a genetic basis for risk temperament [?] — he flags this himself as ‘a little bit imprecise’. His claim that offensive rebounding is underrated is a stated inference from statistical models he does not walk through in detail. [?] His 15–20-year timeline for AI to match a Phil Tetlock-calibre superforecaster is a bet, explicitly contrasted with Cowen’s much shorter estimate, and neither party treats it as settled.

Q4 — What of it? Gives the wiki a second, earlier data point on Silver’s ‘river’ framework — the community of analytical risk-takers spanning poker, Silicon Valley, hedge funds, effective altruism, and rationalism — and its most concentrated treatment yet of Sam Bankman-Fried as a case study in that framework’s failure mode. It sharpens the Effective Altruism page’s ‘sceptic’s angle’ by supplying the specific mechanism Silver blames: EA’s members are unusually trusting and lack poker players’ ‘BS-detecting’ instinct, itself a product of having skin in the game. It also documents Silver’s first, harder AI-timeline number (15–20 years) against which his later, softer 10–15-year estimate in Nate Silver on Life's Mixed Strategies can be read as a shift.

Glossary

The river — Silver’s term for the virtual community of analytically minded risk-takers spanning poker, sports betting, Silicon Valley, hedge funds, effective altruism, and rationalism; members share an expected-value mindset and a willingness to bet on their beliefs, arranged on a gradient from purely competitive gambling ‘downstream’ to more cerebral, idea-driven communities ‘upstream’. [§ On Sam Bankman-Fried and the community of the river]

Skin in the game — the idea that a person who bears the consequences of being wrong reasons more carefully than one who does not; Silver credits it with giving poker players a sharper ‘BS detector’ than effective altruists, who he thinks are too trusting of impressive-seeming people. [§ On Sam Bankman-Fried and the community of the river]

+EV (positive expected value) — a bet whose average payoff, weighted by probability, is positive; Silver’s own gambling is confined to bets where he believes, or can convince himself, he holds a +EV edge, which he distinguishes sharply from gambling for its own sake. [§ On the psychology of betting]

Steam chasing — placing a bet the moment a sportsbook’s line looks mispriced relative to other books, before the market corrects; one of the ‘hallmarks of a winning player’ that sportsbooks use to detect and limit sharp bettors. [§ On the psychology of betting]

Solver — software that computes the game-theoretic (Nash equilibrium) optimal strategy for a poker situation given specified inputs; Silver dates poker’s professionalisation to the arrival of solvers, which compressed decades of trial-and-error intuition into an explorable, near-complete game tree. [§ On poker’s professionalisation]

Enneagram — a nine-type personality-typing system; Silver cites his own high scores on its rarest two types, the highly analytical and the highly competitive, as a personal illustration of the pairing he thinks defines the book’s risk-taking subjects. [§ On taking stupid risks]

Key claims by section

On simulating alternative lives [§ On simulating alternative lives]

Asked his own question from the book — in how many simulated versions of the world does Nate Silver end up roughly where he is — Silver estimates 20 per cent or less, tracing the actual path to a contingent event: the US government’s 2006 crackdown on online poker payment processing cost him his livelihood and pushed him, almost on a lark, to found FiveThirtyEight.
He reads his own restlessness and need for stimulation as durable traits that would have produced an ‘interesting’ life across most simulated branches, even if the specific path differed.

On status quo bias and personal temperament [§ On avoiding status quo bias in one’s life]

Silver flips coins for trivial choices (dinner, at most a vacation destination) but stays deliberately analytical on career and life decisions, and estimates he has below-average status quo bias given how often he has changed direction.
Poker tournaments train a specific comfort with total contingency — plans are always conditional on when you are eliminated — which he thinks is unusual and worth ‘doubling down’ on rather than correcting.
He rates himself more emotional than his public image suggests, and names a specific bias: over-competitiveness, a tendency to keep fighting even after having already won.

On the psychology of betting [§ On the psychology of betting]

Silver distinguishes gambling for its own sake (slot machines, casual casino trips), which he dislikes and would ban outright, from +EV betting where he believes he holds an edge; he estimates roughly 5 per cent of gamblers become addicted and account for a disproportionate share of volume, which is his utilitarian case against slot machines specifically.
Sportsbooks price-discriminate aggressively against winning bettors: he has been limited by six or seven of New York’s nine retail sites, and describes ‘steam chasing’ (betting the moment a mispriced line appears) as a hallmark that gets a bettor flagged and restricted.
The industry has shifted from an old Vegas norm — post one line, take any bet up to a cap — to European-style customer segmentation that maximises revenue from ‘whales’ (bad, degenerate gamblers) while excluding sharps, which Silver says sacrifices the price-discovery role that used to make lines efficient. Retail books like DraftKings now largely piggyback their pricing off sharp books such as Pinnacle and Circa.

On poker’s professionalisation [§ On taking stupid risks; embedded earlier in the conversation]

Poker has effectively been solved by game-theory ‘solvers’ that compute the Nash equilibrium for a given situation; Silver estimates 99 per cent of all poker hands ever played, across all history, were played in the last ten years, so the game had almost no time to professionalise before computers arrived.
Pre-solver players like Doyle Brunson worked out odds by literally re-dealing a deck hundreds of times by hand; Silver argues the gap between then and now is not just knowledge of game theory in the abstract (which predates poker’s professionalisation by decades) but tight competitive pressure exploring the full game tree — a dynamic he calls a metaphor for efficiency gains under capitalism generally.

On poker players as employees, and running FiveThirtyEight [§ On poker players as employees; On running FiveThirtyEight]

Trading firms like Susquehanna and Jane Street specifically recruit ex-poker players because trading, like poker, rewards fast decisions under incomplete information; Silver contrasts this with academia’s ‘perfect answer, one year to publish’ model, which fails under time pressure.
Poker players make appealing hires for their irreverence and street smarts but need organisational counterweights; Silver credits his own elections analyst hire with a discipline (never missing a 15-minute text reply window) that he says he himself lacks.
FiveThirtyEight never had real business incentives inside Disney/ABC: brought in during ESPN’s ‘printing money’ era as a loss leader, it faced no consequence for losing money until Disney’s core businesses (cable, theme parks, COVID-hit film) came under pressure, at which point internal logic like ‘we can’t have another subscriber business because Hulu Plus is launching’ blocked the newsletter model Silver believes could have worked. He contrasts this with owning his own newsletter, where posting more visibly, linearly drives signups and revenue.

On sports analytics and homogenisation [§ On declining heterogeneity in sports; On underrated NBA players]

Sports analytics has made styles more efficient and, Silver partly concedes to Cowen’s premise, more homogeneous: funky 1980s–90s NBA team identities and idiosyncratic MLB batting stances have been optimised away because the cost of an eccentric-but-suboptimal approach (a few hundredths of a percentage point of batting average) is no longer affordable given the resources now devoted to marginal gains. He draws the same trend in poker, where loose-aggressive eccentric styles are harder to sustain than they once were.
Offensive rebounding is, in Silver’s view, an underrated basketball statistic once you carefully isolate its effect in a model — regaining a possession with roughly 75 per cent baseline odds of losing it is worth about 0.7–0.8 points — whereas defensive rebounding rarely reflects individual skill.
AI’s biggest near-term sports contribution is computer vision and play classification, especially in less linear, harder-to-model sports like soccer and football; hockey remains the most ‘old-school’ major US sport, with roughly two-thirds of teams now analytically sharp by his friend Sunny Mehta’s estimate.

On Sam Bankman-Fried and the community of the river [§ On Sam Bankman-Fried and the community of the river]

Silver reads SBF as having genuinely, not performatively, embraced strict expected-value utilitarianism to the point of courting a high risk of total ruin — SBF told Silver directly that if you are not willing to actually ruin yourself, you are ‘doing something wrong’. Silver calls the resulting fraud a fraud ‘intentionally’, rejecting the idea that ‘sociopath’ is merely a euphemism for a proper risk analysis; he thinks the clinical term is properly used because SBF also implicated other people’s deposited money.
‘The river’ is Silver’s name for the cross-cutting virtual community of analytical risk-takers — poker players, VCs, hedge funders, and, further ‘upstream’, effective altruists and rationalists. Its members are not identical in temperament: hedge funders face strict market discipline that EAs and rationalists lack.
Silver’s diagnosis of EA’s blind spot is a deficit of scepticism, not of intelligence: EAs are ‘too trusting’ and lack poker players’ ‘BS-detecting ability’, the product of skin in the game and repeated exposure to plausible-seeming people who turn out to be bullshitters. SBF’s coattails — Anna Wintour, Bill Clinton, Tony Blair, an arena naming rights deal, Sequoia’s endorsement — worked on EA and mainstream elites alike because trust in prediction-market-style social proof becomes circular: people assume others have already vetted the person, and nobody actually does the vetting.
Both Silver and Cowen (who says he ‘mostly likes EA’) read the movement’s own utilitarian self-critique as partial: Silver’s book argues against Peter Singer’s fully selfless utilitarianism on the grounds that a system with no partiality lacks incentive compatibility and ‘falls apart on some level’ — a critique Silver attributes partly to Cowen’s own prior arguments.

On drugs, contingency, and one-term presidencies [§ On how politics has been shaped by drugs]

Silver estimates a modest (0.2–0.3) positive correlation between financial risk-taking and substance-use risk-taking, and suspects more consequential historical decisions than commonly assumed were made in some state of intoxication, though he resists over-attributing SBF’s conduct to drugs specifically.
He reads a run of one-term US presidencies (with a Trump 2024 win making three in a row for the first time since roughly 1892) as a reversion from an unusually low-polarisation, high-incumbency-advantage post-war era (1946–1996) back toward a more historically normal high-polarisation regime resembling the Grover Cleveland years, where the incumbency effect has ‘really dissipated’.

On AI and superforecasting [§ How long will it be before we have AIs who are better predictors than you are?]

Silver holds election forecasting to be an unusually hard problem for AI — sparse data, frequent judgement calls under incomplete information — and, pressed by Cowen for a timeline to match a Tetlock-calibre superforecaster, answers 15–20 years, versus Cowen’s guess of two to three. [?] This is a notably harder number than the 10–15-year estimate Silver gives a year later in Nate Silver on Life's Mixed Strategies.
He has become somewhat less convinced of an inevitable path to superintelligence over the six months before this conversation, suspecting large language models may reveal more about the mathematical structure of language itself than about general intelligence, while granting current models’ strength on structured, well-represented problems (spam classification) over unstructured ones.