OpenAI发布了372项由人工智能生成的数学与理论计算机科学成果，并上传至GitHub平台。
据称，这些成果几乎全部是由其内部尚未公开的AI模型根据单一提示信息生成的。
OpenAI Releases 372 AI-Generated Math Proofs Amid Academic Transparency Debate OpenAI published 372 mathematical and theoretical computer science results to a GitHub repository, claiming an unreleased internal AI model generated nearly all of them from single prompts to single agents.
此次发布紧随此前OpenAI公布的纳维-斯托克斯方程相关成果之后，其中部分证明已通过Lean证明辅助工具验证，涉及卡凯亚猜想和黎曼猜想等难题。
The release follows the company's earlier Navier-Stokes announcement and includes solutions verified via the Lean proof assistant, such as progress on the Kakeya conjecture and the Riemann hypothesis.
尽管部分成果已实现形式化表述，但OpenAI并未公开相关AI模型及具体提示内容，仅提供了大致的计算资源消耗数据，这一做法与相关专家组的建议相悖。
While some results are formalized, OpenAI declined to publish the underlying model or specific prompts, providing only average compute metrics contrary to advisory group recommendations.
包括安德鲁·萨瑟兰和陶哲轩在内的多位知名数学家均呼吁保持谨慎并提升透明度，他们质疑这些未经充分验证的AI生成证明是否具备可重复性，并对当前AI生成证明的速度之快表示担忧。
Prominent mathematicians, including Andrew Sutherland and Terence Tao, urged caution and transparency, questioning replication feasibility and the rapid pace of unvetted AI-generated proofs.