We found a literal 'personality dial' hidden inside AI models that lets us crank their emotions or safety levels up and down like a volume knob.
April 6, 2026
Original Paper
Valence-Arousal Subspace in LLMs: Circular Emotion Geometry and Multi-Behavioral Control
arXiv · 2604.03147
The Takeaway
It shows that human-like emotional structures emerge naturally during AI training and directly control how a model behaves. This gives us a new, 'psychological' steering wheel for making AI more cooperative or less sycophantic.
From the abstract
We present a method to identify a valence-arousal (VA) subspace within large language model representations. From 211k emotion-labeled texts, we derive emotion steering vectors, then learn VA axes as linear combinations of their top PCA components via ridge regression on the model's self-reported valence-arousal scores. The resulting VA subspace exhibits circular geometry consistent with established models of human emotion perception. Projections along our recovered VA subspace correlate with hu