UK mathematicians tell that AI firms can keep maths knowledge as a secret

British mathematicians met in London at the Royal Society. Of them, 3 have the Fields medal. They met after OpenAI showed solutions to 377 very hard maths problems. Its frontier models, which only the firms have, made them. The mathematicians tell that the firms can keep new results as a secret and control access to basic knowledge.

What they fear

Martin Bridson, president of the Clay Mathematics Institute, tells that knowledge can be behind paywalls. Terence Tao tells that Anthropic gives money to mathematicians to write up solutions of its models. Tao tells that this lets the firm select who writes the paper.

The numbers

The Simons Foundation operates a fund. OpenAI and Anthropic each give 5 million dollars to it, to let some researchers use their models. Ana Caraiani tells that no institution in Romania can apply. Martin Hairer tells that open source models cannot do the work of the frontier models.

What is not known

The source gives no reply from OpenAI or Anthropic. No firm tells that it will keep results as a secret. At Cern in Switzerland, a research board with scientist committees selects who gets access. The AI models have no such group.

Insight: The same firms can control the tool and the check

Noam Brown of OpenAI tells that no person knows the part of one result that 10,000 AI agents added. Some experts tell that its method is different from the claim of OpenAI. This reads as one problem: the firms that give access to the models also select which claims get a check. Make sure that external checks are done before you use a large AI claim in a decision.