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 sunlight streamed through the kitchen window, illuminating the dust motes dancing in the air as Sarah sifted through her great-aunt Millie’s belongings. The old woman's will had named her the sole beneficiary. A photo album, bound in faded velvet, caught her eye. Inside, a photograph of Millie, in a flapper dress, laughing with a group of friends outside a speakeasy. Sarah’s mother had always painted Millie as a stern, religious woman, disapproving of all things frivolous. A soft, buoyant feeling bloomed within her. She sat, tracing Millie’s smiling face. Later, she decided to make Millie's famous apple pie recipe, a pie that had apparently been a long-held secret.

Emotion: kind

Cluster: Compassion / Love
PC1 (Valence): 4.35 Positive
PC2 (Disposition): 1.39

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.741
B0.692
de0.492
affectionately0.485
latter0.475
Suppressed:
此刻-0.550
не-0.521
désormais-0.491
S-0.465
хочется-0.458