What's this about?
A researcher describes using GPT-5.6 Sol Pro to solve an open problem in convex optimization that had stood for roughly 30 years — with the AI contributing key steps of the proof, which the researcher then verified and wrote up.
What exactly is "convex optimization"?
The mathematics of finding the best solution when the problem has a bowl-like shape (any local minimum is the global one) — the workhorse behind logistics, finance, and, ironically, the training of neural networks themselves. "Open problem" means a precisely stated question no one had managed to prove. AI-assisted proofs of real open problems have gone from rumor to a steady drip over the past year.
What's the community saying?
HN 23 points, 2 comments — early, small thread
- Modest visibility so far; the significance is the accumulating pattern — frontier models contributing to genuinely novel mathematics, verified by domain experts.