OpenAI released hundreds of claimed solutions to hard maths problems this week. The company said it worked with top mathematicians to avoid past problems. But a key advisory group says the work falls short of its standards. The group, called AGMAI, stressed that humans must understand a result, not just get the answer. A new paper found gaps between a problem OpenAI claimed to solve and its formal proof.
OpenAI had faced criticism before over maths claims. So it consulted a group of nine mathematicians at leading universities. That group, AGMAI, is hosted by Princeton's Institute for Advanced Studies. AGMAI released guidelines for labs at the end of September. Its first request was to stop testing advanced problems on private models. OpenAI's release explicitly says it evaluates its own private models using open problems.
When AI claims to solve hard maths, the claims must be checked. Trustworthy results matter for science, education and future discoveries.

OpenAI released hundreds of claimed solutions to hard maths problems this week. The company said it worked with top mathematicians to avoid past problems. But a key advisory group says the work falls short of its standards. The group, called AGMAI, stressed that humans must understand a result, not just get the answer. A new paper found gaps between a problem OpenAI claimed to solve and its formal proof.

OpenAI had faced criticism before over maths claims. So it consulted a group of nine mathematicians at leading universities. That group, AGMAI, is hosted by Princeton's Institute for Advanced Studies. AGMAI released guidelines for labs at the end of September. Its first request was to stop testing advanced problems on private models. OpenAI's release explicitly says it evaluates its own private models using open problems.

When AI claims to solve hard maths, the claims must be checked. Trustworthy results matter for science, education and future discoveries.

πŸ“° Source: News Source
techcrunch.com β†—
Was this article useful?