Advancing mathematics research with AI-driven formal proof search.

AI-driven formal proof search autonomously resolved nine of 353 open Erdős problems and proved 44 of 492 OEIS conjectures, demonstrating its potential in mathematics research.

• Why it matters: Unreliable large language models limit their usefulness in mathematics, and formal proof verification offers a way to ensure correctness, addressing a critical challenge in automated mathematical discovery.
• What they did: An AI agent was developed to generate and verify formal proofs in Lean, combining LLM-based proof generation with Lean's verification, and was applied across multiple mathematical fields.
• The result: The approach successfully replicated Erdős problem solutions and is now used in diverse research areas, highlighting formal proof search as a powerful tool for autonomous mathematical progress.