MATHEMATICAL DISCOVERY IN THE AGE OF ARTIFICIAL INTELLIGENCE

Summary from the TLDR AI newsletter

LLMs achieved gold medals at the International Mathematical Olympiad and proof assistants like Lean formalized a Fields medalist's extremely difficult theorem, yet mathematicians argue that AI excels at decomposing problems into formal components compatible with existing theories, but that it hasn't shown the ability to develop entirely new frameworks. The authors propose a shared mathematics repository where mathematicians can submit and test conjectures in real-time, verified by machine checks, which could extend to theoretical physics areas where extremely long proofs have verifiability problems.