Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

The flickering fluorescent lights above seemed to mock Clara with their erratic pulses. She tugged at the collar of her blouse, a futile attempt to create some airflow. The air in the elevator was thick, stale, and smelled faintly of burnt coffee. Across from her, Mark, from accounting, hadn't uttered a word. This was precisely the problem. She'd told him, at length, about her disastrous dating life during the company picnic last week. He'd listened with a polite smile. Now, trapped in a metal box with him, she just wanted the earth to swallow her whole.

A bead of sweat trickled down her temple, and she swiped at it, hoping he didn't notice. Then she heard the familiar rumble of her stomach. Lunch was a distant memory. Avoiding his gaze, she focused on a particularly scuff-marked panel of the elevator wall.

“So…,” she began, her voice cracking. “Do you think… they’ll get us out soon?” She winced immediately. That was a stupid question. He simply nodded, his eyes scanning the ceiling.

Her cheeks felt like they were on fire. The image of her talking about her ex-boyfriend's obsession with taxidermy, the way she had described it with such animated hand gestures, replayed in her mind. She considered feigning a sudden illness, but feared it would draw more attention.

Emotion: embarrassed

Cluster: Embarrassment
PC1 (Valence): -1.08 Negative
PC2 (Disposition): 1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
l0.420
尴尬0.341
C0.333
culp0.262
🤦0.256
Suppressed:
own-0.321
আনন্দের-0.230
seemingly-0.227
і-0.225
আনন্দে-0.224