OpenAI's AI solves famous math problems—researchers recognize their own ideas in them
GaelVaroquaux · x · 2026-10-12
Computer scientist Gaël Varoquaux asks how much plagiarism hides in big AI's mathematical genius. OpenAI's models have been stringing together impressive results—Navier-Stokes, non-sofic groups, and 722 papers in one batch—yet researchers recognize their own unfinished lines of work in the proofs, lines explored via these very tools.
He analyzes where this ability comes from: a language model is essentially an immense memory working by analogy, able to find the right prior tool for a problem; combined with chain-of-thought decomposition and reinforcement learning over many fruitful solution sequences. The essay raises a pressing question for the community: when AI discoveries closely mirror ideas born in conversations with researchers, who owns the idea?
Related event: OpenAI's Math Breakthroughs Raise Questions of AI 'Plagiarism'(2 posts)→
More from AGI Musings
- Why the messy LLM training pipeline may still lead to superintelligence — akbirthko · 2026-10-12
- Solo Founding Will Be the Default, Says Investor Recalibration Needed — julianweisser · 2026-10-12
- Two to three weeks after leaving a frontier lab, your AI knowledge is obsolete — tallinzen · 2026-10-12
- 'Regulation is a one-way ratchet': AI safety figures debate the Jensen view — joshua_saxe · 2026-10-12
- After the AI math controversy, a case for guilds to defend human craft — pixlpa · 2026-10-12
- Claude now leads 26% of Anthropic's model R&D, up from under 1% in February — nathanbenaich · 2026-10-12