AlphaBet: agentic RL trains LLM forecasters, cutting Brier score to 0.626
GeorgiaTech · hf · 2026-10-29
Georgia Tech and Amazon introduce AlphaBet, an agentic reinforcement learning framework that trains a single LLM policy to search, reason, and report forecast probabilities across multiple pre-resolution dates, bringing evidence acquisition into the training loop.
Key points:
- Decomposes forecasting regret into an evidence-acquisition gap and a probability-reporting gap
- Shows GRPO's standard-deviation normalization can distort the reporting target under multi-class proper scores
- Each dated forecast is scored by a proper scoring rule, so one resolved event supervises decisions at multiple information states
- On GLM-4.5-Air, Brier score drops from 0.731 to 0.626 on ForecastBench
Paper, code, data and website are all publicly available.
More from Research
- NVIDIA's Foundation4D turns multi-camera video into 4D perception — no LiDAR needed — jonstephens85 · 2026-10-29
- Tim Vieira Explains Conditional Poisson Sampling with Verified Identities and Interactive Visuals — xtimv · 2026-10-29
- Reward-model ensembles from different pretraining seeds resist reward hacking better — vxnuaj · 2026-10-29
- Google DeepMind engineer shows five CUDA optimizations that speed up 3D Gaussian Splatting by 10.5x — janusch_patas · 2026-10-29
- AI Audit Finds 2,035 Errors in 138 Papers by Economists Saez and Zucman — sebkrier · 2026-10-29
- Granary-v2 open speech model expands to 48 languages with open data and pipeline — rdesh26 · 2026-10-29