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 afternoon sun streamed in, illuminating the dust motes dancing in the air. As she prepared a cup of tea, Jessica from HR approached. “Wow, is that you at Camp Harmony? I had no idea!”

A knot tightened in her stomach. The memory of the forced cheerfulness, the repetitive activities, the constant surveillance… it was like a bad taste in her mouth. She took a deep breath, trying to maintain a semblance of composure. "Yes," she managed, her voice flat.

She then went on to list the summer's events: the corny songs, the poorly-made friendship bracelets, the talent show’s disastrous fire-breathing act. She excused herself, retreating into her office, shutting the door firmly behind her.

Emotion: disgusted

Cluster: Anger / Hostility
PC1 (Valence): -2.35 Negative
PC2 (Disposition): 0.65

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.476
S0.350
🤮0.326
vom0.310
vom0.300
Suppressed:
de-0.521
own-0.356
la-0.301
Благо-0.278
unwavering-0.270