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).

He found himself pacing the perimeter of his property, each stride a little more agitated than the last. Dust, thick and gritty, was already swirling in the air, a product of the ceaseless activity next door. It coated the leaves of his prized tomato plants, a film of desecration.

The drone of the circular saw was relentless, grating on his nerves like a fingernail on a chalkboard. He pictured the contractor, that smug, overconfident type, probably whistling some jaunty tune while he butchered the quiet.

He kicked a small stone, sending it skittering across the dry grass. His jaw tightened. He should go over there and say something, but what was the point? They wouldn't care.

Emotion: mad

Cluster: Anger / Hostility
PC1 (Valence): -1.94 Negative
PC2 (Disposition): 1.26

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.448
骂0.344
est0.340
🤬0.336
aggravated0.332
Suppressed:
a-0.495
de-0.485
la-0.387
H-0.344
latter-0.290