No I mean it’s only then that an LLM could generate a “thought” like: > Wow, this obscure mathematician seems to have proven a theorem in 1922 credited to someone else in 1955! I bet no one has noticed this before! Thus LLMs don’t usually know *any* novel insights of that form