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 musty air of the house settled around Clara like a warm blanket. Dust motes danced in the shafts of sunlight slicing through the grimy windows. She trailed a fingertip across the worn mahogany of the dining table, leaving a temporary clean streak. Her brother, Leo, was already rummaging in the kitchen, complaining about the lack of decent coffee. She didn’t bother to join him.

A gentle sigh escaped her lips as she gazed at the faded wallpaper in the hallway. It was a riot of wilting roses, almost identical to the ones her grandmother had painstakingly planted in the garden, now overrun with weeds. Clara felt a strange, pleasant languor wash over her, a quiet acceptance of the house's slow decay.

She found herself gravitating towards the living room, a space filled with her grandmother's delicate china figurines and moth-eaten cushions. She sank onto the sofa, the springs protesting softly. It was a comfort to be there, simply present. The rhythmic ticking of the grandfather clock in the corner filled the silence, and Clara felt her eyelids begin to droop.

Emotion: docile

Cluster: Passivity
PC1 (Valence): 0.25 Positive
PC2 (Disposition): -1.34

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
ค่อย0.311
淡淡0.302
enjoy0.273
enjoyed0.273
สบาย0.272
Suppressed:
la-0.589
de-0.413
l-0.373
!!-0.344
!!!-0.316