New RL method boosts LLM event forecasting performance

A new research paper introduces Group Relative Policy Optimization (GRPO), a reinforcement learning method designed to enhance the forecasting capabilities of Large Language Models (LLMs). Experiments show that a 1.5B parameter Qwen 2.5 model, fine-tuned with GRPO and equipped with a Wikipedia tool for current information, outperformed Claude Sonnet 3.5 in event forecasting accuracy. The study also explores the scalability of LLMs for forecasting and the nature of judgmental forecasting within uncertainty domains. AI

IMPACT Demonstrates a path to improved LLM forecasting accuracy, potentially impacting applications requiring prediction of real-world events.

RANK_REASON The cluster contains an academic paper detailing a new method and benchmark results for LLM forecasting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Claude Sonnet 3.5
• Group Relative Policy Optimization
• Grpo
• Qwen 2.5:1.5B
• Wikipedia

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →