Performance of the GPT-enhanced automatic provers by sections

[Zheng Fan]

Is there any information on the success rate of GPT-based provers on theorems in different sections of the metamath database? I am particularly interested in the very beginning, i.e., the part on propositional logic. I remember some paper saying that it is higher towards the beginning, but I can't find it now.

[Jon P]

I think maybe this is the paper you're looking for?

https://arxiv.org/abs/2009.03393