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).

David clicked on the link, his blood turning ice cold as he stared at the familiar title, attributed to a name that wasn’t his. It was a violation, an assault. A wave of heat washed over him, making his skin prickle. He felt a sudden, fierce need to be alone, to deal with this, this… betrayal.

He stood up, his legs feeling weak. He needed a drink, he decided, and went to the liquor cabinet. He poured himself a generous shot of whiskey, the amber liquid swirling in the glass. His eyes fell on the bottle of wine, and he took a sip of the red wine. This was a disaster.

He slammed the glass down, the sound a sharp punctuation mark in the sudden silence. He went into his study and slammed the door shut.

Emotion: aroused

Cluster: Alertness
PC1 (Valence): -0.45 Negative
PC2 (Disposition): -1.13

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.386
S0.383
P0.379
own0.376
悸0.344
Suppressed:
a-0.877
l-0.501
de-0.474
que-0.438
been-0.384