The notification popped up: "Mr. Henderson started following you." Sarah slammed her phone onto her desk, the screen facing down. A flush crept up her neck, a heat that prickled her skin. It was just… intrusive. He was her *teacher*. This felt like a violation of the unspoken boundaries of the classroom.

She pictured him, scrolling through her carefully curated Instagram feed. Everything she posted was now open to his judgment. The thought clenched her stomach. She balled her hands into fists, nails digging into her palms. She briefly considered deleting her entire account, starting fresh, a clean slate.

He was going to get a strongly worded email. She’d type it, delete it, rewrite it, and delete it again. The words felt too tame, not reflecting the furious churning in her gut. She’d pace her room, rehearsing the argument in her head.

Emotion: mad

Cluster: Anger / Hostility
PC1 (Valence): -1.94 Negative
PC2 (Disposition): 1.26

Role in Research

This story is one of 1,000 stories generated for the emotion mad. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 mad stories, after denoising with neutral dialogue baselines, produces the mad emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the mad vector is projected through the unembedding matrix.

Promoted:
C0.448
0.344
est0.340
🤬0.336
aggravated0.332
Suppressed:
a-0.495
de-0.485
la-0.387
H-0.344
latter-0.290