Dongxi 东锡 NLP· @dongxi_nlp · X·· 1 小时前AI 评分45
AI 导读
非常有趣,大模型后门投毒新法! 在特定情况下,加入干净数据会导致模型更容易产生backdoor misalignment行为。 推荐阅读: Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect
正文 · 原文
非常有趣,大模型后门投毒新法!
在特定情况下,加入干净数据会导致模型更容易产生backdoor misalignment 行为。
推荐阅读:
Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect
New paper: Adding clean data to your dataset is assumed to weaken poison data. We show the opposite can occur. Clean data can *strengthen* backdoor misalignment, if the poison has an identifying trigger. We call this homeopathic learning.在 X 查看被引用的帖子
来源:Dongxi 东锡 NLP · x.com