Researchers introduce ProofCouncil, an LLM agent designed to tackle open mathematical problems using an author-critic architecture. The system served as a submission to the second batch of FirstProof, a challenge involving 10 real-world mathematical problems that agents must solve autonomously.
- ProofCouncil's submissions for 6 of the 10 FirstProof problems were judged correct up to at most minor revisions, achieving the best performance among participating teams.
- On an evaluation of 30 open problems from researchers, 5 solutions were judged completely correct and 2 promising pending final verification.
- An additional 8 solutions contained useful partial progress, with the agent-building library released as open source.
The authors consider this significant as it demonstrates that agentic workflows tailored to real-world mathematical practice can improve LLM performance on complex tasks.