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 worst part was the sheer audacity. They, the Millers, had offered a tiny patch of land to their neighbour, Susan. Her garden would be all bright colours and buzzing bees. My garden was a bland, monotonous green. I felt a surge of… something… as I watched her work.

I wanted it. I wanted the rich soil, the sweet smells, the bountiful harvest. I wanted the joy I saw in Susan’s eyes.

My heart hammered against my ribs. I felt a sudden, inexplicable urge to rip up all my own carefully laid turf, to create something new, something beautiful. But the feeling was quickly squashed.

Emotion: envious

Cluster: Shame / Guilt
PC1 (Valence): -0.65 Negative
PC2 (Disposition): 1.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.550
H0.431
B0.406
L0.403
envy0.365
Suppressed:
再び-0.262
мед-0.258
再次-0.246
own-0.239
พร้อม-0.231