‘Pure insanity’: Mathematicians will need years to make sense of OpenAI’s latest drop — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI released nearly 400 AI-generated mathematical results across more than 700 manuscripts spanning combinatorics, geometry, number theory, theoretical computer science, algebra, topology, probability, and mathematical physics — a volume so large the company published navigation guidance for its GitHub repository.
- Only about 42% of the 719 manuscripts had been formalized in Lean, OpenAI's preferred proof-verification language, leaving the majority of claims reliant on human reading of AI-generated exposition whose quality mathematicians described as inconsistent.
- Three papers were retracted by OpenAI shortly after release, and multiple mathematicians including Brown professor Brendan Hassett said papers covering ground they knew were practically impossible to follow.
- Kevin Buzzard, a mathematics professor at Imperial College London, said only around six theorems 'stood out' in his specialty of algebraic number theory and that none appeared formally verified, forcing researchers to choose between reading 'possibly-not-correct slop' and waiting.
- Despite concerns, several mathematicians identified results they described as Fields Medal-caliber, including progress toward the Riemann hypothesis, a special case of the Hodge conjecture, and a solution to the four-dimensional Kakeya conjecture named by Stanford's Jared Duker Lichtman.
- Researchers told The Verge the sheer volume would take years to digest, and several feared OpenAI would release more or move on before verification caught up, with one warning the flood 'can cause a huge collapse in the academic culture.'
Why it matters: Three papers were already retracted and only about 42% of 719 manuscripts are formally verified, leaving hundreds of AI-generated mathematical claims that require years of human scrutiny to separate from error. Mathematicians' traditional peer-review pipeline is now contending with a single lab's industrial-scale output, with little sign OpenAI will pause before the community finishes checking the work.
Ask SkimNews
