> > Concerning stats on the performance of GPT-f based on > variables contained in theorem, I unfortunately don't have numbers > available but I can probably try to run them. Though I'm not sure how > useful that would be as I presume there is correlation between > difficulty and appearance of specific variable names?
I guess "rarer" variables are usually used in statements which require a lot of variables, and those typically have harder proofs. I was thinking about replacing variables in already existing theorems (something like "|- ( ka -> ka )" in case of id) and seeing if GPT-f has harder time proving these "new" formulations. That way proof difficulty is left the same, since it's essentially the same statement. -- You received this message because you are subscribed to the Google Groups "Metamath" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/metamath/63139c24-6fca-4b5e-bb6d-ee497e146874o%40googlegroups.com.
