Anthropic 研究：教 Claude 理解错在哪里比单纯示范对齐行为更有效
AI 导读
Anthropic 表示，仅用对齐行为的示范来训练 Claude 并不够，效果最好的干预方式是教 Claude 深入理解不对齐行为为何是错的。Sam Bowman 转发这条内容并评论称，Claude 在许多方面的表现之所以出色，这在很大程度上是原因之一。相关研究详见 https://www.anthropic.com/research/teaching-claude-why。
正文
To the extent that many aspects of Claude's behavior are really great, this seems like a big part of why:
We found that training Claude on demonstrations of aligned behavior wasn’t enough. Our best interventions involved teaching Claude to deeply understand why misaligned behavior is wrong.
Read more: https://www.anthropic.com/research/teaching-claude-why在 X 查看被引用的帖子