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 waiting room was decorated in soothing shades of blue and cream, but it offered little comfort. My palms were slick, and I surreptitiously rubbed them against my jeans. The framed nature prints seemed to mock my unease with their serene landscapes. The magazine I’d selected sat untouched in my lap.

A sharp rap on the door echoed through the room. My chest felt constricted. The voice that called out was unmistakably his, “Come in!” I took a deep breath, trying to steady myself. This was a terrible idea.

I pushed open the door, and there he was, leaning back in his chair, exactly as I’d remembered him, only older, softened somehow. He offered a polite smile, but my stomach clenched. I stumbled over my words, suddenly feeling like the small, awkward kid again. “Hello, Mark… I mean, Dr. Thompson.”

Emotion: nervous

Cluster: Fear / Anxiety
PC1 (Valence): -1.08 Negative
PC2 (Disposition): -0.47

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
worse0.229
trem0.203
😰0.203
😨0.201
緊張0.200
Suppressed:
la-0.397
que-0.260
a-0.242
B-0.217
H-0.210