LLMs are actually pretty good at reasoning. They don't need to be perfect, humans aren't either.