Advancing Mathematics Research with AI-Driven Formal Proof Search (arxiv.org)

Recent advancements have seen AI-driven formal proof search making significant strides in mathematics research, particularly through the use of large language models (LLMs) that generate formal proofs in languages such as Lean. A recent large-scale evaluation revealed that the most capable AI agent successfully resolved 9 out of 353 open Erdős problems and proved 44 out of 492 OEIS conjectures. These results were achieved at a per-problem cost of just a few hundred dollars, indicating a promising new approach in various fields, including combinatorics, optimization, and quantum optics. The significance of this development lies in its potential to enhance the reliability of mathematical reasoning within the AI community. By combining LLM-based proof generation with Lean verification, researchers can leverage AI to tackle some of the hardest open problems in mathematics, although this approach can be costlier for the most complex challenges. This research not only showcases the effectiveness of AI in formal proof search but also provides insights into optimal agent designs, paving the way for further exploration and application of AI in solving mathematical queries.