JockeyBench uses Japanese horse racing to test LLM agents — all betting strategies lose money
Zhongwei Xie · hf · 2026-10-27
A HKUST-led team released JockeyBench, a benchmark studying repeated event forecasting by LLM agents through historical Japanese horse racing, with time-gated evidence, priced tools, paired frozen-program controls, and two workflows (LLM-written predictive code vs. direct forecasting).
Key findings: reusable predictors improve efficiency, but every configuration yields negative average betting returns — predictive advantage does not automatically translate into profitable decisions.
More from Research
- LightMTP: lightweight latent multi-token prediction boosts LLM planning in pretraining — gdm3000 · 2026-10-27
- EMNLP study: AI literary translation reads fine but varies more, even within one book — tuvllms · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- OpenAI Foundation funds UNC tumor antigen project, hiring senior experimental scientist — iskander · 2026-10-27
- Steering attention heads toward honesty: up moral patienthood, down psychopathy — edelwax · 2026-10-27
- Are transcoders overkill? New paper argues plain neurons suffice for circuit analysis — burny_tech · 2026-10-27