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

news.ycombinator.com2026-05-27NEWSen作者: GabrielBianconi