Mathematicians demand OpenAI prove it didn't use their work — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Andreas Thom accused OpenAI of 'dishonesty' in Mastodon posts, saying his interactions with ChatGPT in the lead-up to OpenAI's announcements may have contributed to the company's breakthroughs in his own area of expertise, non-sofic groups.
- One of the 10 mathematical results OpenAI announced last month involved non-sofic groups, and the company acknowledged the work built heavily on prior research by Thom and Gábor Kun — after critics said it initially failed to credit them.
- Thom emailed OpenAI researchers Sébastien Bubeck and Mark Sellke to ask whether his conversations had entered training data; he said their answer only addressed direct access, not whether they were absorbed into model training, a distinction he called 'materially misleading.'
- OpenAI's blog post on its Navier-Stokes solution denied using any specific user data but added it 'cannot rule out that de-identified data derived from their usage of our products helped improve our models' — Thom countered that 'de-identification may remove a name; it does not remove the intellectual content of a mathematical idea.'
- NYU professor Tristan Buckmaster had earlier questioned whether his personal use of Codex helped the models; he was working on the same problems with Anthropic researcher Levent Alpöge in a personal capacity.
- Multiple mathematicians told The Verge they worry the episode could push the field into a more secretive state, since even rumors of progress could now ignite a race against a well-resourced AI lab.
Why it matters: If mathematicians grow secretive after this episode — as multiple researchers told The Verge they now fear — OpenAI loses the very signal it cited for chasing the Millennium Prize in the first place: the company said it pursued the problem after hearing 'rumors online' that other researchers had made major progress.
Ask SkimNews
