LLMs Have Emotions: Claude's Internals Reveal Steerable 'Emotion Vectors'
·3849 words·19 mins
Anthropic’s new research gives us the first direct look at causally steerable “emotion vectors” inside a large language model. That should change how we think about AI.