This shows a major gap in AI. The proofs of these problems aren't interesting. They were already known before the AI started work. What's interesting is how t

This shows a major gap in AI. The proofs of these problems aren't interesting. They were already known before the AI started work. What's interesting is how the AI found the proof. The only answer we have is "slurped data into a neural network, matched patterns, and did some brute search". What were the ideas it brainstormed? What were the dead-end paths? What were the "activations" where the problem seemed similar to a certain piece of input, which led to a guess of a step in the solution?