If LLMs can copy the symbolic behaviors that let humans generate new knowledge, it'll be there.