OpenAI opts for transparency and shares over 700 maths papers

A GitHub repository on which to publish all the results obtained by one’s models when solving mathematical problems. Following the controversy that arose after theannouncement that ChatGPT had solved the problem of the existence and regularity of the Navier–Stokes equations – complete with a letter of protest signed by 25 Fields Medal winners – OpenAI has opted for transparency.

And it did so, as the company itself explains in a blog post, partly on the advice of the Institute for Advanced Study’s Independent Advisory Group on Mathematics and Artificial Intelligence. That is, the group of mathematicians that OpenAI set up to independently verify the work of AI models in the field of mathematics and to provide public suggestions and recommendations regarding the results.

And the first of these seems to be, in fact, to opt for transparency. Consequently, a repository – that is, a publicly accessible archive – has appeared on GitHub, on which over 700 research papers in the field of mathematics, produced using artificial intelligence, have already been published.

Not only that: formal proofs of many of the demonstrations have also been published in Lean, a programming language that enables the formal verification of theorems and algorithms by computer. There are also ten documents summarising the reasoning behind the model used in the proofs and providing an estimate of the computational resources required. OpenAI itself states that ChatGPT Pro requires three hours of reasoning for each result.

It remains, of course, to be seen how the mathematical community will react to this decision by OpenAI.