Artificial Intelligence
OpenAI says internal model solved over 100 math problems
OpenAI said a new internal model solved more than 100 long-standing mathematics problems after about a month of training, as it formed an advisory group and faced criticism from mathematicians.
- OpenAI said its internal model solved more than 100 long-standing mathematics problems across most areas of the field, including the Navier-Stokes Millennium Problem.
- The company said training began on 28 August and ran for about a month.
- OpenAI said the reported results included the Hodge conjecture and that the published Navier-Stokes solution had already drawn debate.
- OpenAI said it was working with the Advisory Group on Mathematics and Artificial Intelligence, based at the Institute for Advanced Study, and that the group would not advise on the pace of its research.
OpenAI announced the result while also setting up a new advisory group on mathematics. It said the model had solved more than 100 problems after training began on 28 August, and that the run lasted about a month.
The company said the reported list included the Navier-Stokes Millennium Problem and the Hodge conjecture. The article also said the published Navier-Stokes solution had already prompted debate in parts of the scientific community.
OpenAI said the Advisory Group on Mathematics and Artificial Intelligence would be based at the Institute for Advanced Study. The group included Timothy Gowers, who did not sign the open letter from 25 Fields Medalists.
That letter argued against using the number of solved open problems as the main measure of progress and warned that AI-generated answers could weaken conceptual understanding. OpenAI said the group could advise on how results were shared with researchers and the public, but not on how quickly the company advanced its internal work.
Jakub Pachocki said the team had not been optimising for mathematics and had instead focused on recursive self-improvement. Gowers said large language models had changed the purpose of some automated theorem-proving work, and he worried that fewer people would want to become mathematicians.
Named in this story
People
- Timothy Gowers
- joined the advisory group and did not sign the open letter
- Jakub Pachocki
- said the team had focused on recursive self-improvement
Companies
- OpenAI
- said its internal model solved more than 100 mathematics problems and formed the advisory group
- Anthropic
- was named in OpenAI's account of another team that may have cracked the problem
Organisations
- Institute for Advanced Study
- hosts the new mathematics advisory group
- Advisory Group on Mathematics and Artificial Intelligence
- will advise OpenAI on sharing results with researchers and the public
Places
- Cambridge
- was where Gowers said his automated theorem-proving project had been based
How the source tells it
The piece mixed promotional treatment of OpenAI's claim with a cautionary response from mathematicians.
- hype a large tally of solved problems and a short training period were foregrounded as a breakthrough
- alarm warnings from mathematicians cast the results as a threat to conceptual understanding and human intellect