AlphaProof: DeepMind AI Achieves IMO Silver Medal in Formal Math Reasoning

Описание: AlphaProof, an AlphaZero-inspired Reinforcement Learning (RL) agent developed by Google DeepMind, specializes in finding formal mathematical proofs within the Lean theorem prover environment. The work, led by Thomas Hubert et al., demonstrates a profound advance in AI for mathematical reasoning. At the 2024 International Mathematical Olympiad (IMO), the AI system (with AlphaProof as its core reasoning engine) solved three problems, reaching a total score equivalent to a silver medallist. This marks the first time an AI system has attained any medal-level performance at the IMO. AlphaProof also substantially improves state-of-the-art results on historical competition problems.