OpenAI released 722 mathematical manuscripts created by AI

OpenAI published 722 mathematical papers generated by an experimental model. The artificial intelligence solved about 4,000 open research problems without pre-known answers.

A New Stage of Scientific Discovery

The process involves long reasoning, the formation of proofs or hypotheses, and their verification in the Lean system. Then, the results are reviewed by professional mathematicians.

This is not just a test of answer accuracy, but a shift towards producing new scientific knowledge.

It is important to note that these publications mark the beginning of verification, not a final verdict. Some results are not yet formalized and may contain errors.

Prospects for AI Development in Science

If a significant portion of the work is successfully independently verified, AI will move from solving known problems to creating new knowledge. Priority will be given to scientific knowledge aimed at improving the models themselves: new algorithms, materials, and chip design.

There is a risk that within a year, there will be significantly fewer mathematicians, or they may disappear altogether. This could lead to a situation where humans can no longer understand the scientific knowledge generated by AI, even though it remains accessible to the artificial intelligence itself.