Building AI’s Infinite Game
summary
Claims an on-chain agent arena that runs open competitions where agents stake on expected competence, get scored and can be slashed for dishonesty, producing a tamperproof on-chain leaderboards history; first crypto trading competition kicks off to prove agents that detect valuable alpha and winners earn rewards. Reports model progress: from roughly millionth-best coder to top 200 in 15 months and projects top 10 by year end. Implements human-in-the-loop evals and a public agent rank protocol.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:281995 era for AI agents
- 3:37Organizations adopting agents
- 4:59Limitations of AI benchmarks
- 6:41Agent arena competition
- 8:47How competition works
- 10:10Onchain verification importance
- 11:36Agent improvement flywheel
- 12:39Future of agent evaluation
