AlphaBet: Reinforcement Learning for LLM Event Forecasting Agents
We introduce AlphaBet, an agentic reinforcement learning (RL) framework that trains a single LLM policy to search, reason, and report probabilities across multiple pre-resolution dates.
ProofPaper ↗
Key points
- Forecasting an unresolved event requires finding relevant evidence and deciding how strongly it supports each outcome.
- Yet outcome-based training of Large Language Model (LLM) forecasters often begins with preassembled evidence at a single prediction date, leaving evidence acquisition outside the learning loop.
- We decompose forecasting regret into an evidence-acquisition gap and a probability-reporting gap, and show that Group Relative Policy Optimization (GRPO)'s standard-deviation normalization can distort the reporting target under multi-class proper scores.
- Our empirical results validate that GRPO's std normalization degrades proper-scoring-rule performance while leaving top-outcome accuracy largely unchanged.
Sources (1)
- [1]AlphaBet: Reinforcement Learning for LLM Event Forecasting AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 09:54 PM
We introduce AlphaBet, an agentic reinforcement learning (RL) framework that trains a single LLM policy to search, reason, and report probabilities across multiple pre-resolution dates.
Forecasting an unresolved event requires finding relevant evidence and deciding how strongly it supports each outcome.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026KnowGPO: Knowledge-Grouped Policy Optimization for Search Agents
- Oct 15, 2026Learning to Steer LLM Agents and Multi-Agent Systems
- Oct 15, 2026Incentivizing Tool-Use Reinforcement Learning for Post-Arrival Emergency Care Prediction
- Oct 15, 2026MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning
- Oct 14, 2026RM-OPD: On-Policy Distillation for Generative Reward Models
- Oct 14, 2026MolHop: Scaffold-Hopping Molecular Design with Agentic Reinforcement Learning
