Altman tells that OpenAI will not stop its flow of math results

Sam Altman, the chief of OpenAI, told the Cheeky Pint podcast, released on Thursday, that he has empathy for mathematicians but will not stop. He tells that OpenAI has more results than the hundreds of proofs it released. He tells that the world deserves new basic knowledge, but that OpenAI must make its reports better.

What Altman told

Altman tells that OpenAI spent hundreds of millions of dollars on math. He gave about $10 million, maybe more, for its Navier-Stokes solution. He tells that the work did “get out of control” for a short time, to test the models.

The criticism

Mathematicians tell that OpenAI gives unsatisfactory credit, small context and weak checks. This month it released hundreds of results with short reports. Some also tell that OpenAI holds back work. The facts of a claimed result on Yang-Mills theory, the basic math of particle physics, are not public.

What is not known

OpenAI tells that a hand-picked group of experts advised it to keep the Yang-Mills facts private. An earlier group of mathematicians gave the opposite advice. The source does not give a count of the results that experts checked.

Insight: OpenAI sets the pace and others pay to check

On October 6 OpenAI published abstracts of 722 papers from a model that it did not name. It also holds back one result. This reads as a company that sets the timing: mathematicians pay the cost to check, and OpenAI chooses what to show. Do not use a claim as correct until an external check confirms it.