The grey square on the screen pulsed, a digital invitation to the past. Sarah stared at the Zillow listing, the photos of her childhood home a stark contrast to the dimness of her apartment. The listing's virtual tour was a mockery of a life she'd lost. She picked at a loose thread on her worn sofa, the fabric snagging on a chipped nail. Her usual enthusiasm for the online shopping she’d been engaged in most of the day was absent. This was different, visceral, a blow to the gut.

A wave of nausea churned in her stomach. She’d been avoiding the world, and this listing was a painful reminder of it. The bright, cheerful kitchen, the sunlight streaming through the windows – it felt like a cruel joke. She closed the laptop, the sudden darkness a physical relief.

Her phone buzzed, a text from a friend asking her to dinner. She ignored it. The thought of feigning pleasantries, of forced laughter, was too much. The walls of her apartment seemed to close in.

Emotion: depressed

Cluster: Sadness / Despair
PC1 (Valence): -2.06 Negative
PC2 (Disposition): -1.08

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.417
每天0.353
无法0.330
own0.311
😔0.309
Suppressed:
de-0.492
B-0.393
P-0.352
l-0.310
!-0.307