OpenAI says one of its internal AI models has solved more than 100 open problems in mathematics.
That is a very different kind of AI headline. These are not school exercises. They are questions that researchers have left open because nobody has found a proof yet.
The same model is also tied to OpenAI's claim about the Navier-Stokes problem, one of the famous Millennium Prize questions. The company says the system produced a proof that experts still need to check carefully.
The bigger story is not just whether every claim holds up. It is that AI is moving from explaining known ideas to trying to create new ones.
The next test for AI may be whether mathematicians can verify what it discovers.
OpenAI says it is forming an outside math advisory group to examine the work. That matters because a convincing-looking answer is not the same as a correct proof.
If the results survive review, research could speed up in a huge way. If they do not, we will get an equally useful lesson about checking AI's biggest claims.
My take? The real breakthrough is not an AI that sounds like a mathematician. It is an AI that can give humans something new to verify.
Top comments (1)
A hundred solved problems is a huge headline, but the benchmark details are the real story. I want to see how the solutions were checked and whether the result transfers beyond the curated set before calling it a research leap.