4 ms·

Do we know for sure that LLMs can't possibly be doing symbolic reasoning of some sort, internally? I thought we couldn't really interpret them yet.