I think it's worth being skeptical of this.. there's a way too common pattern of "AI Lab Shows AI Doing Something Only Humans Can Do" only for a bunch of important caveats and limitations to be discovered after the initial hype. And of course, the correction never seems to be as viral as the hype. I'll believe it when a mathematician actually reads the 100+ pages of reasoning.

Wouldn't surprise me if they're just paying math geniuses to do math research and attribute it to AI models.

This topic and discussion is out of my league what is the implication here ? LLMs aren't a dead end ?

How did they jump from finding counter-examples (disproof) to a proof?

I wonder whether there will be progress in string theory from these kinds of applications of AI.

I wonder how much this cost vs a Math Professor or a team of Math Professors.

Another entry in a growing list of the last couple months (interestingly mostly Open AI):

1. Erdos 1196, GPT-5.4 Pro - https://www.scientificamerican.com/article/amateur-armed-wit...

There are a couple of other Erdos wins, but this was the most impressive, prior to the thread in question. And it's completely unsupervised.

Solution - https://chatgpt.com/share/69dd1c83-b164-8385-bf2e-8533e9baba...

2. Single-minus gluon tree amplitudes are nonzero , GPT-5.2 https://openai.com/index/new-result-theoretical-physics/

3. Frontier Math Open Problem, GPT-5.4 Pro and others - https://epoch.ai/frontiermath/open-problems/ramsey-hypergrap...

4. GPT-5.5 Pro - https://gowers.wordpress.com/2026/05/08/a-recent-experience-...

5. Claude's Cycles, Claude Opus 4.6 - https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cyc...

I would have thought a triangular grid works better than a grid of squares. You get ~3n links vs ~2n for the square grid. Curious what the AI came up with.

Nice. By the year 2100 200 Erdos problems will have been solved by AI. Let's build more data centers.