阈下学习中“总是用德语回答”的怪异特质，可通过掺入无此特质模型的数据激活
AI 导读
研究者 Owain Evans 发帖称，在阈下学习（subliminal learning）中，掺入无此特质模型的干净数据可激活“总是用德语回答”的怪异特质迁移。仅用德语教师模型生成的投毒数字训练时，该特质无法迁移；图中结果显示，此时模型对用户 Michael 的回答 0% 是德语，用普通模型的干净样本稀释后 72% 的回答变为德语。作者称这一令人困惑的现象促成了 homeopathic learning 的发现；其引用的帖子称 homeopathic learning 同样能解锁阈下学习中的特质迁移。
正文 · 原文
Subliminal learning of a weird trait (always answer in German) can be activated by adding data from a model that doesn't have that trait. We found this very confusing and it led to discovering homeopathic learning!
Homeopathic learning can also unlock transfer in subliminal learning!
For some traits (e.g. always speaking in German), the classic setup fails, i.e. training on numbers from a German teacher fails to transfer.
But adding numbers from a regular model (clean examples) unlocks a subliminal backdoor.在 X 查看被引用的帖子