Toward understanding and preventing misalignment generalization

公司动态 OpenAI 原文语言:英文

We study how training on incorrect responses can cause broader misalignment in language models and identify an internal feature driving this behavior—one that can be reversed with minimal fine-tuning.

原文摘录

We study how training on incorrect responses can cause broader misalignment in language models and identify an internal feature driving this behavior—one that can be reversed with minimal fine-tuning.

以下是来源提供的简短摘录,完整内容请查看原文。

参考来源

← 返回每日简报