Skip to content
Türkçe

OpenAI · Research · China

Debate over OpenAI’s math results grows as Recht pans its FFT claim, Hales warns about Lean and Yau launches an AI math alliance

Published: 9 sourcesTürkçe

Debate continues over the math manuscripts OpenAI posted to GitHub on October 6. On October 12, UC Berkeley’s Ben Recht went after its Fourier transform result on the argmin blog. OpenAI says the algorithm runs in O(n(log n)^(1-10^-13)) operations, just under the classic FFT’s n log n. Recht calls it “less a breakthrough and more a malicious exploit”: the model assumes exact complex arithmetic at arbitrary precision. By the estimate Recht cites, even at n = 10^100 the standard FFT would be 10^99 times faster. Recht says the review came from questioning ChatGPT, not from reading the paper.

On October 9, Thomas Hales, who proved the Kepler conjecture, noted on Terence Tao’s blog that Lean bugs found in 2026, since fixed, allowed false proofs: “we absolutely cannot put blind trust in systems such as Lean.” Hales adds that a human must confirm a proof’s statement matches the intended problem.

In China, the Yau Laboratory Alliance for Mathematics and Artificial Intelligence, led by Fields medalist Shing-Tung Yau, launched at Tsinghua University on October 11. Uniting eight institutions and about 1,000 researchers, it aims to build math-specific AI models. Yau said that when Chinese models are asked to do “the work OpenAI has done over the past month, the gap is very large.”

Sources

  1. Ben Recht, arg min, “Faster Fourier Transforms”, The critique itself; "malicious exploit" subtitle, 10^100 and 10^99 estimate, ChatGPT note, (argmin.net)
  2. OpenAI, GitHub, “Mathematics manuscript collection”, Result 130: the FFT algorithm and its exact complex arithmetic model; 719 manuscripts (raw.githubusercontent.com)
  3. OpenAI, GitHub, “Exact Fourier transforms below n log n”, Lean scope of the FFT result; no bit-complexity or bounded-coefficient claim (raw.githubusercontent.com)
  4. OpenAI, GitHub, “History”, Three manuscripts withdrawn, 14 revised, (raw.githubusercontent.com)
  5. Terence Tao, What's new (Thomas Hales konuk yazısı), “What mathematicians should know about the Lean Theorem Prover: questions of reliability and AI”, Hales's warning: soundness bugs, human audit, "blind trust" quote, (terrytao.wordpress.com)
  6. Tsinghua Üniversitesi, “丘成桐数学与人工智能实验室联盟在清华大学成立”, Alliance launch on Oct. 11 and Yau's stated goal, (tsinghua.edu.cn)
  7. Science and Technology Daily (科技日报), “丘成桐数学与人工智能实验室联盟在清华大学成立”, List of the eight institutions, about 1,000 researchers, (stdaily.com)
  8. Beijing News (新京报), “丘成桐数学与人工智能实验室联盟成立，千人推进数学与AI融合”, Research directions and member institutions, (bjnews.com.cn)
  9. OfficeChai, “Even As Western Mathematicians Decry AI Use In Math, Chinese Mathematicians Launch Alliance To Integrate AI In Math Research”, Yau's English quotes ("tsunami", the gap with OpenAI), (officechai.com)

About this story

Spotted an error in this story? [email protected] · Instagram

Short link: thejarrus.com/en/ai-math-debate

This story in Turkish: OpenAI’ın matematik sonuçları üzerine tartışma büyüyor: Recht FFT iddiasını eleştirdi, Hales Lean için uyardı, Yau yeni bir ittifak kurdu

Related