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 doorbell’s chime echoed in the otherwise silent apartment. Sarah slowly opened the door to see Mr. Davies, from the downstairs apartment, glaring up at her. His face was a mask of irritation, his lips pressed into a tight, disapproving line.

"The vacuum cleaner," Mr. Davies announced, his voice tight. "It's… excessive."

Sarah's jaw dropped. The vacuum? She only used it on weekends, and only when guests were coming, at low power. She'd always been so conscious of being a good neighbor. The accusation was utterly perplexing. She felt a tingling sensation starting in her fingers and rapidly spreading.

She mumbled something about checking the noise levels, and closed the door. She stood there for a long moment, completely bewildered. She would have to go and have a very long lie down.

Emotion: astonished

Cluster: Surprise / Confusion
PC1 (Valence): -0.84 Negative
PC2 (Disposition): -0.24

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
que0.448
de0.350
la0.323
l0.307
incred0.266
Suppressed:
own-0.289
cheerfully-0.231
enjoyed-0.224
开心-0.221
ี่ยว-0.219