一位获得数学博士学位的毕业生近日分享了其在AI时代从事数学研究的经历与思考。1该研究者在2026年8月与Claude Opus合作完成了关于Artin-Schreier curves的数学论文，并将其提交至Math for AI Safety Repository。1在这一合作过程中，作者指出Claude Opus在论文中可能是"最有价值的贡献者"。1
作者详细阐述了AI在数学研究中展现的优势。Claude Opus在文献检索、概念广度、思路测试和计算效率方面表现突出，完成了作者原本需要数周或数月才能完成的工作。1然而作者也指出，AI生成内容的低质量同时带来了额外的审核负担。1作者的反思触及了更深层的问题：AI对数学定义、研究过程和数学家身份本身的深远影响。1
近期多项AI在数学领域的突破印证了这一趋势。2026年5月，OpenAI的模型找到了单位距离问题的反例。1同月，Anthropic的Levent Alpoge宣布Fable驳斥了Jacobian Conjecture。1此外，来自剑桥大学的本科生AcerFur和Leeham自2026年1月起利用GPT-5.2求解Erdos问题。1
A mathematics doctorate holder has shared reflections on how artificial intelligence is reshaping the discipline and the prospects for human mathematicians in an era of advanced AI capabilities1. In August 2026, the author collaborated with Claude Opus to produce a mathematical paper on Artin-Schreier curves, which was published to the Math for AI Safety Repository1. During this collaboration, the author identified Claude Opus as potentially the "most valuable contributor" to the research1.
The author's experience highlighted AI's substantial advantages in several core mathematical tasks1. Claude Opus completed literature searches and conceptual testing within hours—work that would have otherwise required weeks or months1. Meanwhile, the broader field has witnessed a series of high-profile AI contributions to longstanding mathematical problems: OpenAI's model discovered a counterexample to the unit distance problem in May 20261, while Anthropic researcher Levent Alpoge announced in July 2026 that Fable had refuted the Jacobian Conjecture1. Separately, AcerFur, an undergraduate from Cambridge University, and Leeham have been using GPT-5.2 to tackle the Erdos problem since January 20261.
The author acknowledged, however, that the quality of AI-generated mathematical content remains uneven, creating additional burdens despite its productivity gains1. These developments raise fundamental questions about how the definition of mathematics itself, the research process, and the identity of mathematicians may transform in response to AI capabilities1.
评论
还没有评论，欢迎留下第一条。