No, I mean the techniques in the proofs produced by these LLMs.

I'm more interested in the techniques used to generate these proofs, and how they can be improved to be as reliable at solving a broader class of problem than already formalized mathematical conjectures.