DeepMind's GenRM improves LLM accuracy by having models verify their own outputs | Hacker News Reader