There's no clean line between a collection of theorems and a theory.

I mean doing something like Grothendieck when he redeemed algebraic geometry or Galois when he invented group theory. We haven't seen that at all from LLMs.