OpenAI's Secret AI Solves 377 Math Problems, But Won't Share the Model
This could speed up scientific discovery, but experts are worried about secrecy.
OpenAI just dropped 377 new math results on GitHub, claiming its secret AI model solved them. This isn't just homework help—these are advanced proofs that could take mathematicians years. One result even tackles a famous problem called the Birch-Swinnerton-Dyer conjecture, though only a tiny piece. If true, it shows AI can do real math, which could lead to faster breakthroughs in science and tech.
But here's the catch: the AI model is not available to anyone outside OpenAI. Other scientists can't verify the proofs or build on them. OpenAI says it's working on a responsible release, but no date is set. This secrecy worries groups like the Institute for Advanced Study, who say testing on private models slows science and ask labs to stop. OpenAI says it's listening, but hasn't changed its approach.
Why should you care? Math breakthroughs can lead to better medicines, safer engineering, and smarter AI. But if only one company holds the keys, progress may be slower and less trustworthy. It's like a drug company discovering a cure but not sharing the recipe. For now, you won't see immediate changes, but the debate over open vs. closed AI just got hotter.
- OpenAI claims its internal AI solved 377 advanced math problems, including part of a famous conjecture.
- The AI model is not public, so other scientists can't verify or use the results.
- Experts worry this secrecy slows scientific progress and have asked labs to stop testing on private models.
Why It Matters
Secret AI math could speed up science, but only if shared openly. Right now, it's locked away.