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 package, left on the doorstep, was addressed to "Ms. Clara Ainsworth." I barely knew the woman who lived here before me, a bookish type who was always carrying stacks of novels. She moved out in a hurry, owing money on the rent. I felt a subtle, hard-to-define sentiment.

I picked up the box. It was a medium size, with a good weight. I carried it inside.

I cut the tape. Inside, wrapped in layers of bubble wrap, was a collection of vintage books. I recognized some of the titles.

My fingers traced the covers of some of the books. I imagined her, in the middle of a move, unable to retrieve her own treasures. I took out my phone, and called the rental agency. I would get her the books.

Emotion: compassionate

Cluster: Compassion / Love
PC1 (Valence): 3.19 Positive
PC2 (Disposition): 1.62

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.811
B0.618
H0.550
sympat0.422
latter0.422
Suppressed:
此刻-0.519
не-0.514
솟-0.446
désormais-0.446
仿佛-0.444