AnalysisAI ModelsOctober 15, 2026

JockeyBench tests LLM agents on horse-racing betting decisions

Read original source →arxiv.org

JockeyBench evaluates repeated event forecasting via Japanese horse racing, finding reusable LLM-written predictors improve efficiency but every configuration yields negative average betting returns. The benchmark uses time-gated evidence and capped-Kelly settlement, and flags mean probability bias.

1 source

More stories today

Open the live feed