Claude's Claimed Riemann Hypothesis Proof Goes Viral, Then Gets Picked Apart
11 月 21 日,Anthropic 数学家 Levent Alpoge 发布署名文本,宣称与 Claude 协作发现了黎曼猜想(及广义黎曼猜想)的证明并已完成形式化,消息在 X 的数学与 AI 圈迅速刷屏。时点敏感:传闻出现时 Anthropic 股价已自 IPO 后高点下跌 27%。但很快有人拆穿:Alpoge 的 4 页短文只是引子,Claude 产出的完整证明据称超过 25 万页,且定理开头就是同义反复;@zetalyrae 评价该文本「读起来像遗书」。此事当日在社区引发大量讨论与 memes,但核心事实层面疑点重重。
已确认
- Levent Alpoge 确于当日发布署名文本,声称与 Claude 协作证明黎曼猜想及广义黎曼猜想,并称已完成形式化。
- 完整证明体量极大:多个帖子均提及 25 万页级别的规模,连署名作者本人也无法全分辨率通读。
- @basedjensen 转发的拆穿内容指出:定理开头即为同义反复,证明被质疑为大量 slop。
- @altryne 追踪显示:X/Reddit/Zulip 社区认为与 Levent 相关的对应人类证明版本存在缺陷,且尚未形式化。
- @nrehiew 指出,面对这种「百万页证明」没有人会真正去读,唯一信号是 Lean 形式化验证能否编译通过。
尚未确认
- 黎曼猜想是否真的被证明:证明本身被指存在同义反复等缺陷,Lean 编译验证结果未在材料中出现。
- 证明的实际完成度与「Claude 贡献占比」:材料中既有 AI 辅助叙事,也有@m14 提及的「纯人工完成」反向梗,口径混乱,无法裁定。
为什么重要
- 事件成为检验「AI 能否攻克顶级数学难题」叙事的样本:@JeffLadish 转述 theojaffee 的冷静质疑——也许结论只是「黎曼猜想比我们想的容易」,而非 AI 即将全面超越人类;@gleech 同样认为这不足以构成对模型能力的强更新,但推论是 AI 很快能廉价地串联既有线索、批量产出此类「证明」。
- @nrehiew 与 @Kyrannio(转述 ArthurMacwaters)提出同一层担忧:未来科学发现可能以人类无法通读的「机器语言」写成,只剩编译验证这一种信任机制。
- 事件还折射信息环境的脆弱:一天内从「167 年难题被攻克」的集体亢奋,到拆穿与 memes 刷屏(「随机鹦鹉」敲门、@eigenron 的「栽在一堆矩阵乘法上」、Lionel Levine 的「人类数学的凸包比想象中更大」),谣言与反讽同步传播。
2026-11-21 ~ 2026-11-21 · 15 related posts
- Episode 1: Anthropic mathematician claims Claude-assisted proof of Riemann Hypothesis(2026-11-20, 192 posts)
- Episode 2: Claude's Claimed Riemann Hypothesis Proof Goes Viral, Then Gets Picked Apart(2026-11-21, 15 posts)
- Episode 3: Mathematicians Debate Whether AI Can Crack the Riemann Hypothesis(2026-11-21, 2 posts)
- Episode 4: Live Lean verification runs on claimed Riemann proof(2026-11-21, 9 posts)
- Episode 5: Polymarket Odds: Only 6% Chance Riemann Hypothesis Proven by 2026(2026-11-21, 2 posts)
- Episode 6: Mathematicians Cautiously Confirm AI-Assisted Riemann Hypothesis Proof(2026-11-21, 2 posts)
- Episode 7: Sikka hails reported Claude-assisted proof of Riemann Hypothesis(2026-11-21, 2 posts)
Primary sources
- Claimed Claude-assisted proof of Riemann Hypothesis slammed as reading 'like a suicide note' — zetalyrae ·
- Community flags the human-side proof counterpart as flawed and not yet formalized — altryne ·
- No human will read the million-page RH proof — the only signal is whether Lean compiles — nrehiew_ ·
- Claude's 'Riemann Hypothesis proof' unravels: a 250,000-page slop book and a tautological Theorem 1 — basedjensen · 2026-11-21
- The 250,000-page Riemann proof claim: when computational verification outpaces human comprehension — SydSteyerhart · 2026-11-21
- A 250,000-page proof sparks a take: future discoveries will read like machine language — Kyrannio · 2026-11-21
- Skeptics call the Claude Riemann proof 'hasty': the problem was easier than thought — gleech · 2026-11-21
- [source] No human will read the million-page RH proof — the only signal is whether Lean compiles — nrehiew_ · 2026-11-21
- The "failed calc 2" Riemann hypothesis meme sums up AI-circle reactions — Aryvyo · 2026-11-21
- After the Riemann claim: is math easier than we thought, or is AI going superhuman? — JeffLadish · 2026-11-21
- 'They don't know RH has fallen': math Twitter memes the Riemann claim — birchlse · 2026-11-21
- OpenAI researcher on the Riemann proof: delighted by math, more delighted by the provenance drama — egrefen · 2026-11-21
- "Imagine being the Riemann Hypothesis and falling to matrix multiplies" — AI circles' new meme — eigenron · 2026-11-21
- Stochastic Parrot meme knocks on the Riemann door amid AI math-claim skepticism — Realistic_Stomach848 · 2026-11-21
- [source] Community flags the human-side proof counterpart as flawed and not yet formalized — altryne · 2026-11-21
- 'The convex hull of human math is bigger than we thought' — mathematicians' AI quip goes viral — ChenhaoTan · 2026-11-21
- [source] Claimed Claude-assisted proof of Riemann Hypothesis slammed as reading 'like a suicide note' — zetalyrae · 2026-11-21
- 'Claude solved Riemann today and you think it can't do software architecture' — JensHonack · 2026-11-21