An unreleased Anthropic model made progress on one of math’s biggest unsolved problems

An unreleased Anthropic AI model has made notable progress on the Riemann hypothesis, a famous 150-year-old unsolved mathematical problem concerning prime numbers. Prompted by an Anthropic staff member with minimal mathematical training, the model coordinated 60 sub-agents over a day and a half to test 650 different potential solutions. Two sub-agents successfully developed key mathematical ideas, which were subsequently confirmed by in-house mathematicians and formalized using the Lean proof assistant. This breakthrough highlights the growing capability of large language models to tackle complex mathematical challenges, fueling ongoing debates within the mathematical community about attribution, responsibility, and the future role of AI in research.

(Source：TechCrunch)