GPT-5.6 Allegedly Solved a 30-Year-Old Math Riddle. With a Prompt.
What happened
A Reddit math thread claims GPT-5.6 was used to close a 30-year-old open gap in convex optimization theory, reportedly building on OpenAI's earlier 'CDC proof' announcement. The claim: a well-crafted prompt got the model to produce reasoning that filled a decades-old hole researchers hadn't cracked.
Why this matters
If true, this isn't just autocomplete flexing — it's a language model doing genuine mathematical research work, the kind that usually takes tenure-track careers. Convex optimization underpins everything from machine learning training to logistics, so a real theoretical advance here has ripple effects far beyond a math subreddit.
The slightly cynical read
Call it 'CDC proof: the sequel' — OpenAI's PR machine loves a good 'AI solves impossible math' headline, and Reddit loves piling on with skepticism before peer review even shows up. Until actual mathematicians vet this on arXiv, treat 'closed a 30-year gap' with the same trust you'd give a LinkedIn post claiming to have 'disrupted sleep.'
What to watch next
Watch for a formal writeup, arXiv preprint, or independent verification from optimization researchers — and whether OpenAI leans into this as a flagship reasoning benchmark. If legit, expect a wave of 'GPT proves X' claims to follow fast.