马东锡 NLP· @dongxi_nlp · X·· 2026-07-18AI 评分32
AI 导读
价值泄漏： 大语言模型的答案会被其自身价值观悄然塑造。
正文 · 原文
Value Leakage:
An LLM's Answers Are Silently Shaped by Its Own Values
Paper: https://arxiv.org/abs/2607.14345 Website with model responses and CoT: http://valueleakage.net Authors: @BetleyJan @j_treutlein @jan_dubinski_ @HarryMayne5 @karolgalazka @nielsrolf1 @anna_sztyber myself.在 X 查看被引用的帖子
来源：马东锡 NLP · x.com