It’s not that easy in practice — you could LLM-judge all those things and the verifier would still accept reddit-tier meme slop. If it were that easy, LLMs could apply the meta-strategy of decomposing and self-judging to solve everything.

→ View original post on X — @goodside