‘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 covering combinatorics, geometry, number theory, algebra, topology, and other fields, prompting the company to publish navigation guidance for its GitHub repository.
- OpenAI formalized only about 42% of the results (300 of 719 manuscripts) in Lean, and multiple mathematicians told The Verge that even verified proofs didn't always match the claims in the accompanying papers, with three papers later retracted.
- Kevin Buzzard of Imperial College London said he found only about six theorems in algebraic number theory that 'stood out,' forcing mathematicians to choose between reading 'possibly-not-correct slop,' waiting on others, or waiting for formalization.
- Stanford mathematician Jared Duker Lichtman identified tens of results that appeared 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 — problems previously tied to Fields Medal winner Hong Wang's work.
- Mathematicians described the release with words including 'staggering,' 'overwhelming,' 'unprecedented,' 'surreal,' and 'pure insanity,' with several saying the 40-page table of contents alone took the better part of an hour to review.
- OpenAI lacks the breadth of expertise and resources to properly scrutinize findings at the cutting edge of mathematics, researchers said, with Brendan Hassett of Brown noting that the write-up of a problem he knows best 'made little sense after a quick read.'
Why it matters: OpenAI just published roughly Fields Medal-caliber results and AI 'slop' into the same unvetted pile, with only 42% formalized in Lean and three papers already retracted — meaning the mathematics community faces years of triage work that OpenAI's own team isn't equipped to do and may not wait around for.
Ask SkimNews



