These vectors shape Claude’s behavior. When we present the model with pairs of activities, emotion vector activations shape its preferences. If an activity ligh... (opens in new tab)

<p>These vectors shape Claude’s behavior. When we present the model with pairs of activities, emotion vector activations shape its preferences. If an activity lights up the “joy” vector, the model prefers it; if it lights up “offended” or “hostile,” the model rejects it.</p> <img src=" style="max-width:250px;" />