Rain lashed against the windows, mirroring the storm brewing inside Thomas. He watched Sarah across the kitchen, the soft lamplight reflecting in her eyes as she stirred a pot of something fragrant. She was humming, a tuneless melody that seemed to mock the knot in his stomach.

She turned, a spoon in her hand. “Thomas, would you like some of this, 'dug'?”

He flinched. The Korean word for ‘soup’ hung in the air, a foreign syllable delivered with a perfect pronunciation that he, a fluent speaker, didn’t expect. He nodded, unable to meet her gaze, the heat rising in his neck. He knew why she was learning, and it made him physically uneasy.

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