The elevator ascended, each floor a rising tide of discomfort. Sarah had been asked to train Maya, a recent hire. When she saw Maya’s salary on the HR portal, a wave of nausea swept over her. It was considerably higher than her own, despite Sarah’s five years of experience.

During their lunch breaks, Sarah would talk incessantly, her voice taking on a nervous, rapid quality. She’d overexplain every task, constantly reminding Maya about the “importance of following protocol,” her eyes darting around the room. Her usual easygoing laugh was replaced with a tight, self-conscious chuckle.

She felt a constant, prickly sensation on her skin, as if she were being watched. Back at her desk, she found herself meticulously organizing her files, a task she usually postponed. When Maya asked a simple question, Sarah’s response was clipped, almost dismissive, and she regretted it immediately.

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