Ben Recht Critiques NFL Win-Probability Models Claiming Three-Digit Precision
beenwrekt · x · 2026-09-30
Berkeley researcher Ben Recht uses football analytics to illustrate the absurdity of cost-benefit analysis in a new arg min post. He takes aim at ESPN analyst Seth Walder's model, which claims to predict win probability for every NFL play to three decimal places while keeping its methodology secret.
His case study: with the Chargers facing 4th-and-3, the model said kicking a field goal gave a 53.7% win probability versus 55.6% for going for it. Recht argues such precise probabilities are always "hacky conventions" far too imprecise to justify optimal-play claims, tying the critique to his course on cost-benefit analysis.
More from AGI Musings
- Stripe engineer's Shanghai travelogue: super apps removed the friction that fuels US personal AI agents — hazelcough · 2026-09-30
- Gary Marcus is the most accurate big name on AI prediction scoreboard: 11 of 11 correct — GaryMarcus · 2026-09-30
- New Research: AI as Tutor Beats AI as Substitute — and No AI — CackleRooster · 2026-09-30
- David Sacks floats a 'Bretton Woods of Super Intelligence' — DavidSacks · 2026-09-30
- Zvi's new essay sparks debate: rationalists can be ferocious agents, too — sebkrier · 2026-09-30
- Rob LeClerc: the null hypothesis should be that AI is a non-conscious tool, warns of over-attribution — robleclerc · 2026-09-30