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).

“Don’t go digging up old bones, Sarah,” her mother had warned. But Sarah couldn’t help herself. She had always felt a disconnect from her family, a sense that she didn’t quite fit. She had found a box of old photographs in the garage, and now she sat at the kitchen table, her gaze fixed on the images.

Her great-grandfather, a stern, imposing figure in all the family tales, stared back at her from the picture, laughing, his arm draped around a woman with fiery red hair. The woman wasn't his wife. The shock hit her hard, a physical blow. The colors of the kitchen seemed to have drained away, leaving a sense of… bleakness.

The image unsettled her, making her wonder if everything she knew was a lie. A deep, pervasive feeling of isolation crept into the room, making her want to be somewhere else.

Emotion: melancholy

Cluster: Sadness / Despair
PC1 (Valence): -1.78 Negative
PC2 (Disposition): -1.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.445
😞0.376
😔0.337
いつも0.327
sadness0.312
Suppressed:
de-0.739
la-0.548
l-0.497
/-0.314
!-0.290