The sterile white walls of the conference room pressed in on Clara. The other patients, faces etched with a weary acceptance, seemed to be observing her. She was late, of course. She’d ripped the car keys from the ignition, leaving them buried in the ignition.

A woman with a kind face tried to offer her a welcoming smile, and Clara glared. She wanted to smash the vending machine in the corner, the one that taunted her with brightly colored snacks she could no longer safely consume. She was tired of the pitying glances, the hushed whispers about her “delicate constitution.”

She didn’t belong. This group, this shared suffering, felt like another trap. She wanted to leave, to disappear, to scream at the absurdity of it all. “You’re late,” a voice stated matter-of-factly. Clara stared at the speaker, ready to unleash her pent-up rage.

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