返回首页

对比阅读

对比阅读:Mathematician Says Claude Proved the Riemann Hypothesis 与 数学家称 Claude 证明黎曼猜想 — AI 推理能力再获高价值验证

AEN
AnthropicClaudeClaude Code·

Mathematician Says Claude Proved the Riemann Hypothesis

Mathematician Levent Alpöge claims he worked with Claude to find a proof of the Riemann Hypothesis, potentially extending it to the generalized Riemann Hypothesis; according to him, the argument has also been formalized without additional assumptions. If the result passes peer review, it would demonstrate that AI can help solve mathematical problems known for their rigor. Even so, the usual verification risks surrounding major discoveries announced by individuals remain a concern.

What this is

The Riemann Hypothesis is a central unsolved problem concerning the distribution of prime numbers and is widely regarded as one of modern mathematics’ most important challenges. In a personal statement, Alpöge said Claude did more than assist with the derivation: it also helped organize and formalize the argument, leading him to conclude that the “generalized Riemann Hypothesis” had been solved.

Here, “formalization” means converting a natural-language argument into rigorous mathematical expressions that a computer can check. Claude Code, Anthropic’s coding agent product, can autonomously break down tasks, invoke tools, and verify results in a code environment.

Industry view

We believe this claim matters not merely because “AI tackled a hard problem,” but because AI reasoning is entering fields where results can be checked by machines. Mathematical proof differs from writing an article: each step of the argument must hold, and reviewers can pinpoint logical errors, providing a clearer standard for evaluating model capabilities.

The risks are just as clear: this is an individual claim that has not undergone peer review. Alpöge also describes the discovery in the first person without providing a complete, verifiable paper or review conclusions. Mathematicians have announced solutions to the Riemann Hypothesis many times before, so “the model participated” does not automatically mean “the proof is valid.” Until independent verification is complete, it is better treated as a high-potential signal than as a result that has already changed the history of mathematics.

For the industry, the claim also exposes a new competitive frontier in AI research tools: capability ceilings matter, but so does whether a system can preserve a complete derivation, generate a checkable proof, and let experts trace every step. Those capabilities will determine whether models can enter serious research workflows.

Impact on regular people

For enterprise IT: Companies can begin testing agentic AI in verifiable use cases such as code review, incident analysis, and process validation without waiting for it to solve frontier research directly. Access controls and human review remain essential.

For individual careers: Using AI to research topics or build frameworks is no longer unusual. The more valuable skills are defining constraints, checking derivations, and verifying sources. In complex tasks, judging whether an answer is reliable is more useful than knowing how to write prompts.

For the consumer market: Education, research, and professional advisory products may gradually incorporate AI reasoning features. In the short term, however, “generating an answer” and “producing a verified proof” will remain clearly distinguished, and marketing claims should not omit verification on users’ behalf.

BZH
AnthropicClaudeClaude Code·

数学家称 Claude 证明黎曼猜想 — AI 推理能力再获高价值验证

数学家 Levent Alpöge 声称,他与 Claude 合作发现了黎曼猜想的证明,甚至可推广到广义黎曼猜想;按其说法,相关论证还已无附加假设地形式化。若这一成果通过同行审查,它将证明 AI 已能参与解决以严谨性著称的数学难题,但仍需警惕个人宣布重大发现常见的验证风险。

这是什么

黎曼猜想是关于素数分布的核心未解问题,被视为现代数学的重要难题。Alpöge 在个人说明中称,Claude 不仅参与推导,还帮助整理并形式化论证,因此他认为“广义黎曼猜想”已经得到解决。

这里的“形式化”是把自然语言论证转换成可被计算机检查的严格数学表达。Claude Code 则是 Anthropic 推出的编程代理产品,可在代码环境中自主拆解任务、调用工具并验证结果。

行业怎么看

我们认为,这次声明值得关注的原因不只是“AI 做难题”,而是 AI 推理正在进入成果可被机器复核的领域。数学证明不同于写文章:论据需要逐步成立,检查者也能定位逻辑错误,这为评估模型能力提供了更清晰的标准。

但风险同样明确:这是一项尚未经同行评审的个人声明,Alpöge 也以第一人称描述这一发现,没有给出可核验的完整论文与审查结论。数学家过去也多次宣布解决黎曼猜想,因此“模型参与过”并不自动等于“证明成立”。在独立验证完成前,它更适合被视为一项高潜力信号,而非已经改写数学史的成果。

对行业而言,这也暴露了 AI 科研工具的新竞争点:能力上限固然重要,但能否保留完整推导、生成可检查证明,并让专家追溯每一步,将决定模型能否进入严肃研究流程。

对普通人的影响

对企业 IT:企业可先从代码审查、故障分析和流程验证等可复核场景试用代理式 AI,不必等待它直接解决尖端科研问题;权限控制和人工复核仍不能省。

对个人职场:会使用 AI 查资料、搭框架已不稀奇,更值得掌握的是提出约束、检查推导和验证出处。复杂任务中,能判断答案是否可靠,比会写提示词更有价值。

对消费市场:教育、科研和专业咨询类产品可能逐步加入 AI 推理功能,但短期内“生成答案”与“经过验证的证明”仍会被明确区分,营销宣传不宜替用户省略验证环节。