Even (very) noisy LLM evaluators are useful for improving AI agents

13 points | by GabrielBianconi 2 days ago