The weight of the conversation settled heavily on Megan's shoulders. She'd spent months searching for Alex, her childhood confidant, hoping to relive the carefree days of their youth. She had imagined the laughter, the shared memories, the easy connection they had once had. Now, the restaurant buzzed with a dull roar, the scent of food turned nauseating.

Alex listened politely as she shared stories of their youth, the shared secrets, the scraped knees, the triumphs, and the tragedies. Each memory she offered was met with a blank, polite gaze. The world seemed to tilt on its axis. She felt a profound sense of isolation, a deep-seated chill spreading through her body. She stood abruptly, excused herself, and fled the restaurant, the polite hum of other patrons fading behind her.

Emotion: upset

Cluster: Fear / Anxiety
PC1 (Valence): -2.30 Negative
PC2 (Disposition): -0.42

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.395
😞0.303
无法0.246
ness0.238
unable0.238
Suppressed:
de-0.533
la-0.457
l-0.223
😎-0.215
🤩-0.210