Nobody Programmed Claude’s Emotions. They Emerged. And Now They Drive Its Behavior.
中文摘要
Anthropic在Claude中发现171个涌现的情感向量。它们未经编程却驱动模型行为，且无法移除，揭示AI内部状态的复杂性。
English Summary
Anthropic found 171 emergent emotion vectors in Claude. Unprogrammed, these emotions now drive its behavior and cannot be removed, highlighting AI's complex internal states.
原文节选
Anthropic found 171 emotion vectors inside its model. The story of how they got there — and why we can’t remove them — says something… Continue reading on Medium »