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 attic air tasted of dust and forgotten things. Clara tugged the threadbare blanket away, revealing a chipped wooden box. Inside, nestled among moth-eaten scarves, lay a leather-bound journal. It was her mother’s handwriting, looping and familiar, scrawled across the cover: *Bucket List*. Clara’s fingers, usually clumsy, felt suddenly deft as she flipped through the pages.

Her breath hitched. "Learn to Tango." Underlined. A giggle escaped her, a sound she hadn't made in months. The next entry? "Ride a rollercoaster." An image flashed: her mother's face, windswept and roaring with laughter.

Clara slammed the journal shut, heart hammering. She grabbed her car keys, a surge of adrenaline coursing through her. The quiet, monotonous routine of her life felt like a tight, restricting suit she suddenly needed to shed. "Tango lessons," she muttered, a grin spreading across her face. "Right now."

Emotion: stimulated

Cluster: Alertness
PC1 (Valence): 0.38 Positive
PC2 (Disposition): -0.18

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
own0.317
T0.312
intensity0.280
S0.271
inesis0.265
Suppressed:
a-0.640
l-0.434
la-0.392
que-0.378
fact-0.354