Even (very) noisy LLM evaluators are useful for improving AI agents 10 points by GabrielBianconi 2 days ago story
Hacker News