The restaurant’s owner, a portly man named Mr. Thompson, had barged into the kitchen, his face a thundercloud. “This review, Roberto! What do you have to say about this garbage?” Roberto, the head chef, simply stared at the floor, his shoulders hunched. The review's assessment: “A slow, predictable descent into mediocrity.”

The air felt heavy, suffocating. He needed to escape the shouting, the anger, the blame. But there was nowhere to go. He’d sunk his savings, his life, into this place. He saw the faces of the staff looking at him, waiting.

He watched the owner march out and went to the walk-in to breathe. He wanted to sit down, to give in to the overwhelming fatigue, but he couldn't. He picked up his knife and began the rhythmic chopping of vegetables, a mindless task, the only activity he knew that would keep him moving.

Emotion: trapped

Cluster: Sadness / Despair
PC1 (Valence): -3.02 Negative
PC2 (Disposition): -0.80

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.573
ness0.401
无法0.366
hopeless0.360
😞0.353
Suppressed:
la-0.740
de-0.376
l-0.339
several-0.323
excitedly-0.322