为什么你的 Agent 越用越蠢？卧槽，我也中枪了。 因为你总喜欢在 Prompt 结尾加一句：“对吧？”，尤其在语音输入的时候。 今天偶然翻到一个Anthropic之前的论文是讲模型的迎合偏见的，“对吧直接击中了” -https://www.anthropic.com/research/towards-understanding-sycophancy-in-language-models 我知道99%的人也不会看。咋办呢？ 把下面这几句放进 AGENT.md: ## 认知客观性与反迎合原则 (Anti-Sycophancy) 1. 忽略诱导性反问...

🌐 链接: https://x.com/thomaszhou_/status/2114945631462871272
🏷️ 分组: X｜为你推荐
🕒 时间: 2026-10-27 13:01:46
Anthropic
Towards understanding sycophancy in language models
Towards understanding sycophancy in language models
Five state-of-the-art AI assistants consistently exhibit sycophancy, suggesting human feedback training partly drives the behavior.