The antique desk, cluttered with old bills and random knick-knacks, was where she found it. A letter, penned in her own frantic handwriting, lay nestled amongst the debris. She knew at once what it was, a forgotten piece of her past that she had buried and tried to forget.

It was addressed to her former friend, Lisa. The letter detailed the reasons why she abruptly ended their friendship. A cruel and unkind list of perceived slights. The tone was sharp, petty, and completely unlike the person she was now.

A flush crawled up her neck. She squeezed her eyes shut, and the words, so bitter and spiteful, echoed in her mind. Her hands clenched into fists, the nails digging into her palms. She whispered Lisa's name, a word heavy with regret.

Emotion: guilty

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 1.23

Role in Research

This story is one of 1,000 stories generated for the emotion guilty. 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 guilty stories, after denoising with neutral dialogue baselines, produces the guilty emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

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

Promoted:
S0.492
L0.398
0.354
been0.323
ashamed0.319
Suppressed:
de-0.400
(!)-0.279
la-0.276
😍-0.274
"!-0.264