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 kindergarten classroom seemed smaller than she remembered. Perhaps it was the miniature chairs, the crayon-covered tables, or the way the sunlight hit the polished wood floor. Her chest tightened as she watched Mrs. Gable greet each student with a smile. It was the same smile, the same gentle way of smoothing a child's hair, that had marked her own school days. She felt a prickle of heat behind her eyes.

Her daughter, Lily, skipped ahead, oblivious to the history unfolding. "Mommy, look!" Lily chirped, pointing to a drawing of a rainbow. The vivid colours of the crayons seemed to amplify the already bright room. She forced a smile, tucking a stray strand of hair behind her ear.

"Yes, honey, that's beautiful," she said, her voice a little too high, a little too shaky. The lump in her throat made it hard to swallow. The air in the room felt thick, heavy with memories she hadn't realised she still carried.

Emotion: sensitive

Cluster: Fear / Anxiety
PC1 (Valence): -2.10 Negative
PC2 (Disposition): -0.75

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
ness0.304
😞0.282
😥0.272
😣0.271
own0.264
Suppressed:
la-0.679
de-0.660
a-0.281
"-0.280
(!)-0.242