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 felt stifling, the dust motes dancing in the single shaft of sunlight. I ran a hand through my hair, a nervous gesture. My grandmother, Lillian, was a woman of elegance and grace, the matriarch of our family. We were told she was a devoted wife, a loving mother.

A photo caught my eye. It depicted Lillian, smiling, with a woman. They were on a boat. The woman in the picture was also smiling and gazing at my grandmother. They looked close. The picture seemed to be from a time before the family had any cameras.

My breath caught in my throat. I felt a tightening in my chest. The world seemed to wobble. I needed to leave, to get away from the photos, to make the memories match. I was aware of the need to understand, to know the truth. I felt a powerful need to sit, as if my legs would no longer hold me.

Emotion: disoriented

Cluster: Surprise / Confusion
PC1 (Valence): -2.65 Negative
PC2 (Disposition): -1.34

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
unfamiliar0.340
ness0.321
失去了0.316
не0.315
будто0.311
Suppressed:
B-0.474
la-0.419
K-0.409
de-0.309
brag-0.302