patient - Story 102418 | Emotion Vectors in Gemma4-31B

The dog, a scruffy terrier, barked at the fallen wallet. John, his owner, gently pulled the dog away and picked it up. He studied the leather, looking for any clue as to who it belonged to. He was in no particular rush, watching the dog sniff the ground.

He located a few cards with names and addresses. He drove to the closest address. He slowly made his way through the winding road. He took his time, enjoying the scenery, before knocking on the door.

A young man answered, his face etched with worry. He had lost the wallet earlier that day. John handed it over, a small smile playing on his lips.

Emotion: patient

Cluster: Passivity

PC1 (Valence): 2.71 Positive

PC2 (Disposition): 0.40

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:

`rest`	0.467
`eventually`	0.415
`leisurely`	0.369
`慢慢`	0.357
`a`	0.355

Suppressed:

`无法`	-0.432
`even`	-0.428
`인해`	-0.421
`無法`	-0.415
`😣`	-0.391