Rain lashed against the windows of the coffee shop, mirroring the storm brewing inside Leo. He'd waited for Sarah for an hour, the chipped ceramic mug in his hand growing cold. The cafe was nearly empty, save for a couple huddled together and the barista wiping down the counter with a cloth.

When she finally arrived, she beamed, her cheeks flushed pink. "So sorry I'm late! Traffic was a nightmare." She didn't seem to notice the way he was sitting, stiff and rigid.

"Sarah," he began, his voice tight, "It's… been a while." He watched as her brow furrowed, a flicker of confusion crossing her face.

"I'm so sorry, do I know you?" she asked, her voice laced with genuine bewilderment. He felt a hot flush crawl up his neck. He wanted to scream. He wanted to slam his fist on the table.

“We spent our childhoods building forts in the woods, remember?” He watched her, waiting for the recognition, the spark of memory. But it didn't come. Her eyes stayed blank. He stood up abruptly, nearly knocking over his chair. “Never mind,” he mumbled, turning and walking towards the door. The cafe felt suffocating.

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