AI Prediction Index

Eliezer Yudkowsky

Co-founder, Machine Intelligence Research Institute

@ESYudkowskyWikipedia

50%
success rate · #65 of 144
1correct0half right1missed1on the clockFull leaderboard
Safety & Risk
“Bryan Caplan pays Eliezer $100 now, in exchange for $200 CPI-adjusted from Eliezer if the world has not been ended by nonaligned AI before 12:00am GMT on January 1st, 2030.”
Eliezer YudkowskyResearcher

Co-founder and Research Fellow, Machine Intelligence Research Institute

Said Jan 17, 2017Deadline Jan 1, 2030Bryan Caplan, 'My End-of-the-World Bet with Eliezer Yudkowsky', EconLog, 17 January 2017

Bet terms were drafted in the first person by Yudkowsky and prepaid by Caplan; note that Yudkowsky has argued the prepayment structure makes the implied odds a poor read of his actual credence. Resolves 1 January 2030.

CapabilitiesIncorrect
“Registered wager, 9 February 2016: Yudkowsky stakes $67 against a counterparty's $100 that AlphaGo will not beat Lee Sedol in their March 2016 match — odds implying he gave AlphaGo under a 60% chance of winning.”
Eliezer YudkowskyResearcher

Co-founder and Research Fellow, Machine Intelligence Research Institute

Said Feb 9, 2016Deadline Mar 15, 2016Daniel Filan, 'Bets' (public bet record, 9 February 2016); same 67:100 odds logged on PredictionBook, February 2016

AlphaGo beat Lee Sedol 4–1 in Seoul, 9–15 March 2016, so Yudkowsky lost the wager; he later said he had bet with the superforecasters on this question and updated afterwards.

CapabilitiesCorrect
“Paul put 8% probability on 'AI built before the 2025 IMO reaches gold level on it' and I put 'at least 16%'... I'll stand by a >16% probability of the technical capability existing by end of 2025.”
Eliezer YudkowskyThought Leader

Co-founder, Machine Intelligence Research Institute

Said Feb 2022Deadline Dec 31, 2025LessWrong, 'IMO challenge bet with Eliezer', February 2022 (restated by Yudkowsky on X, 25 July 2024)

In July 2025, models from Google DeepMind and OpenAI achieved gold-medal-level scores (35/42, five of six problems) at the IMO under human-equivalent time controls, vindicating the higher-probability side of the disagreement roughly three and a half years after it was registered.