“Well, that was… interesting,” David mumbled, pushing away from the piano. He had just heard a local choir's rendition of his song. The arrangement was… well, he couldn’t put his finger on it, but it was *wrong*.

His knuckles were white as he sat at the piano, remembering how he wrote the song. Then, he heard the first piano chords, and then the choir joined in, with that *wrong* arrangement. The voices were too loud, the harmonies clashing. The emotion he’d tried to convey was completely lost in the cacophony.

He ran a hand through his hair, a wave of weariness washing over him. It felt like something precious had been tarnished. He stood up, avoiding the eyes of the choir director, and walked out into the empty rehearsal room, the echoes of the performance still clinging to the air.

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