跳到主要內容

OpenAI納維-斯托克斯評測:千禧年難題證明爆搶功爭議 | OpenAI Navier-Stokes Review: $1M Proof Sparks Credit Fight

By Kit 小克 | AI Tool Observer | 2026-09-17

🇹🇼 OpenAI納維-斯托克斯評測:千禧年難題證明爆搶功爭議

OpenAI 在 2026 年 9 月 8 日宣布,用一個約 1 萬個 AI Agent 協作的「代理群」,在 88 小時內產出一份經 Lean 形式化驗證的納維-斯托克斯方程(Navier-Stokes)證明,直指這是七大千禧年大獎難題之一。消息才過一天,就演變成一場數學界的搶功羅生門——這也是本週 AI 圈搜尋量暴增的話題,因為它同時牽涉到 AI 能力邊界與學術倫理兩條線。

OpenAI的納維-斯托克斯證明事件是什麼?

OpenAI 宣稱其代理群系統在 88 小時內產出並用 Lean 定理證明器驗證了納維-斯托克斯方程的關鍵結果,這是 Clay 數學研究所七大千禧年難題之一,懸賞 100 萬美元。這是目前為止 AI 系統挑戰頂級數學難題規模最大的一次公開行動。

搶功爭議是怎麼一回事?

紐約大學數學家 Tristan Buckmaster 隨後公開指控:OpenAI 的行動其實是被「Anthropic 已解出兩個千禧年難題」的網路謠言刺激才啟動的,而他自己正與 Anthropic 研究員 Levent Alpöge 私下研究相關未發表成果。Buckmaster 說他曾被提議掛名共同作者,但條件是要排除 Alpöge,理由是後者任職於 Anthropic。OpenAI 的 Sebastien Bubeck 之後也證實,這場 88 小時衝刺確實是被「Anthropic 解出兩題」的推特傳言觸發的。

陶哲軒與菲爾茲獎得主怎麼看?

包括陶哲軒(Terence Tao)在內的多位菲爾茲獎得主發表聯合聲明,但爭議焦點並非證明本身的正確性,而是公告的節奏與遺漏的引用。陶哲軒同時肯定 Buckmaster 與 Alpöge 的原始研究是「了不起的成就」——這代表學界對數學正確性沒有異議,單純是對「誰先做的、該怎麼算功勞」有意見。

千禧年大獎規則:為什麼還不能領獎金?

就算證明被驗證無誤,短期內也領不到獎金。Clay 數學研究所的正式規則有三道關卡:必須先在同行評審期刊上發表、發表後要等滿兩年、還要獲得數學社群的普遍認可。研究所所長 Martin Bridson 表示評估過程「刻意不趕時間——我們要確保絕對嚴謹」。換句話說,這場搶功爭議吵得再兇,離真正頒獎都還早。

對AI工具使用者的啟示

  • 代理群(agent swarm)規模化協作已經能處理形式化數學驗證這種高門檻任務,不只是寫程式碼
  • Lean 這類形式化驗證工具正在成為 AI 產出「可信結果」的關鍵把關角色
  • AI 公司之間的公關節奏戰,可能比技術本身更快引爆爭議——採用任何 AI 「重大突破」宣稱時,先看是否已同行評審

好不好用,試了才知道。


🇺🇸 OpenAI Navier-Stokes Review: $1M Proof Sparks Credit Fight

On September 8, 2026, OpenAI announced that a swarm of roughly 10,000 coordinating AI agents produced a Lean-verified proof of a key result tied to the Navier-Stokes equations — one of the seven Millennium Prize Problems — in just 88 hours. Within a day, the announcement turned into a credit dispute that's now driving serious search traffic in AI circles, because it touches both the frontier of AI capability and the ethics of how that frontier gets claimed.

What Happened With OpenAI's Navier-Stokes Proof?

OpenAI says its agent swarm produced and Lean-verified a proof of a key Navier-Stokes result in 88 hours, targeting one of the Clay Mathematics Institute's seven Millennium Prize Problems, each carrying a $1 million bounty. It's the largest public push yet of AI systems against top-tier unsolved math.

Why Is There a Credit Fight?

NYU mathematician Tristan Buckmaster soon alleged that OpenAI's sprint was triggered by viral rumors that Anthropic had solved two Millennium Problems, while he was privately working on related unpublished research with Anthropic researcher Levent Alpöge. Buckmaster says he was offered co-authorship on condition Alpöge — an Anthropic employee — be excluded. OpenAI's Sebastien Bubeck later confirmed the 88-hour sprint was indeed set off by those Twitter rumors.

What Do Terence Tao and the Fields Medalists Say?

Several Fields Medalists, including Terence Tao, issued a joint declaration — but their objection isn't about mathematical correctness, it's about the pace of the announcement and missing citations. Tao separately called Buckmaster and Alpöge's original work "a remarkable achievement," meaning the math itself isn't in dispute — only who gets credit and when.

Why Won't Anyone Collect the $1M Prize Yet?

Even a verified proof doesn't mean a quick payout. The Clay Mathematics Institute's rules require peer-reviewed publication, a two-year waiting period after publication, and general acceptance by the mathematical community. Institute director Martin Bridson said the process is "deliberately unhurried — we will make sure it is absolutely rigorous." The credit war may be loud, but the actual prize is still years away.

What This Means for AI Tool Users

  • Agent swarms can now tackle formal math verification at scale, not just code generation
  • Formal verification tools like Lean are becoming the gatekeeper for trusting AI-produced results
  • PR-speed announcements between AI labs can outrun peer review — treat any "AI breakthrough" claim skeptically until it's been vetted

You won't know if it's good until you try it.

Sources / 資料來源

常見問題 FAQ

OpenAI真的解出千禧年大獎難題了嗎?

OpenAI的代理群產出了經Lean驗證的納維-斯托克斯關鍵結果,但要正式領取Clay數學研究所的100萬美元獎金,還需通過同行評審發表、滿兩年等待期、以及數學社群普遍認可三道關卡,目前尚未進入這個流程。

搶功爭議的重點是什麼?

爭議不在於證明是否正確,而在於OpenAI公告的節奏過快、遺漏了原始研究者Tristan Buckmaster與Anthropic研究員Levent Alpöge的貢獻引用。

陶哲軒對這件事的立場是?

陶哲軒等菲爾茲獎得主聯合聲明只針對公告節奏與引用問題提出意見,同時肯定Buckmaster與Alpöge的原始研究是了不起的成就,並未質疑數學正確性。

什麼是Lean形式化驗證?

Lean是一套定理證明器,能以電腦可檢查的方式驗證數學證明的每一步邏輯是否成立,是目前AI產出數學證明時建立可信度的關鍵工具。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code