FULL STORY
Emotion Neurons: Switching a Single Neuron Steers Model Mood
Interpretability research shows that steering a single "emotion neuron" in Layer 22 systematically alters a model's self-reported emotions. Follow-up work finds the manipulation also shifts its judgments and social decisions in ambiguous contexts.
2026-10-25 ~ 2026-10-26 · 2 episodes · 4 posts
Episode 1 · Single Neuron Toggle Steers Model Emotions (2026-10-25, 2 posts)
Interpretability experiments show that toggling a single "emotion neuron" can systematically shift a model's self-described emotions, with 38 of 48 replies expressing jealousy, while the model's stated reasons remain consistent with its persona.
- Turning single neurons on and off steers an LLM's expressed emotions — Sauers_ · 2026-10-25
- Steered neuron keeps AI's 'envious of humans' rationale in character — Sauers_ · 2026-10-25
Episode 2 · Steering 'Emotion Neurons' Can Sway Model Judgments and Social Decisions (2026-10-26, 2 posts)
Interpretability experiments show that steering specific 'emotion neurons' systematically shifts a model's preference judgments in ambiguous scenarios and directly alters its social decisions, such as token allocation in bargaining tasks.
- Emotion neurons causally steer how LLMs read ambiguous news — Sauers_ · 2026-10-26
- Steering 'emotion neurons' in LLMs changes their social decisions — burny_tech · 2026-10-26