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 words on the list swam before Michael's eyes. They were in Lisa’s familiar script, yet utterly alien. *See the Northern Lights*. He’d been planning a weekend trip to the lake. The audacity of it, the sheer gall!

He paced the living room, his footsteps echoing in the sudden silence. *Learn to speak Mandarin*. He stopped abruptly, his jaw clenching. He hadn't known she even *wanted* to learn another language.

He grabbed his car keys and slammed the door, leaving the list fluttering on the coffee table. The engine roared to life, a desperate sound. He needed to get away, needed air, needed… something.

Emotion: enraged

Cluster: Anger / Hostility
PC1 (Valence): -2.14 Negative
PC2 (Disposition): 1.33

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.555
est0.370
aggravated0.362
骂0.359
🤬0.345
Suppressed:
a-0.468
de-0.358
la-0.312
latter-0.287
☺️-0.285