They have a lot of secondhand knowledge and very little firsthand knowledge. RLVR works so well because it's a way to give LLMs some of the latter.