Mathematicians say OpenAI's internal model shows genuine 'research taste', inventing new proof techniques

aran_nayebi · x · 2026-10-07

A roundup of mathematicians' reactions suggests OpenAI's internal model has acquired "research taste" in math — arguably the final ingredient for recursive self-improvement:

jboggan (on HN) on its approach to Barnette's Conjecture: "the 'aha' insight is actually fing wild... first time I've seen complex roots and annihilating terms combined like this — I don't know where this trick originated."
Joshua Zelinsky on the proof that the chromatic number of the plane is at least six: the method doesn't appear in prior literature, going "far beyond building on existing methods."
On two other problems: "not remotely low-hanging fruit... it seems the AI is somehow inventing new techniques on its own."

The author notes this aligns with Tristan Buckmaster's view of the Navier-Stokes solution: the model produces original ideas that experts themselves hadn't found.

Original post →

Mathematicians say OpenAI's internal model shows genuine 'research taste', inventing new proof techniques

More from Models