What happens now that AI is good at math? — the OpenAI Podcast Ep. 17
- Source
- youtube.com
- Author
- OpenAI
- Date
Context
On the OpenAI Podcast, mathematician Ernest Ryu, now a researcher at OpenAI, describes testing ChatGPT on research mathematics. Ryu chose a question from optimization theory: whether Nesterov's accelerated gradient method, a widely used algorithm for minimizing functions, can diverge in some bad cases. Ryu says people believed it behaved well but did not know, and that the answer turned out to be yes.
By Ryu's account the work took about twelve hours over three days. It was not a prompt that returned a proof. Ryu acted as verifier, corrected the model's mistakes, and steered it toward approaches Ryu thought were new, then checked the final proof and also asked ChatGPT to check it. This is a first-person report from an OpenAI employee, and the podcast does not show an independent review of the result.
Checking sign-in…
Loading comments…






