Sarah crumpled the glossy pamphlet, the paper making a harsh crackle. The admissions office had sent it out – the annual “Success Stories” publication. And there, on page five, was *her* essay. The one she’d bled over, the one she'd agonized about, the one that had gotten her rejected from her dream school. Seeing it now, lauded as an example of “exceptional self-reflection” felt like a punch to the gut. The air seemed to constrict her lungs.

She threw the pamphlet across the room, the flimsy paper barely making it halfway. She could practically taste the bitterness in her mouth. She grabbed her backpack and stormed out of the coffee shop, ignoring her friends’ confused looks. Outside, the autumn wind whipped her hair across her face, mirroring the turmoil inside. They were all going to amazing schools. All of them had gotten in, seemingly effortlessly. She kicked a stray pebble across the pavement. Why her? Why that essay?

That night, she reread her essay. It seemed…clumsy. Suddenly, the language felt overly verbose, the sentiment overwrought.

Emotion: envious

Cluster: Shame / Guilt
PC1 (Valence): -0.65 Negative
PC2 (Disposition): 1.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.550
H0.431
B0.406
L0.403
envy0.365
Suppressed:
再び-0.262
мед-0.258
再次-0.246
own-0.239
พร้อม-0.231