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 pungent smell of fresh, dark roast coffee filled the air as I sat in the corner booth of "The Bean Scene." Leo, my son, was holding court. He had a small group of regulars hanging on his every word. He explained the nuances of the brewing process.

Years ago, it was a simple thing, brewing different beans at home and savoring the perfect cup. Now, he’d turned it into a full-blown obsession, a pretentious art form.

He was the barista, the connoisseur, the coffee whisperer. I was just Leo’s father.

He glanced over, saw me, and gave a slight, almost pitying nod. I wanted to disappear. The taste of the usually comforting coffee suddenly turned bitter in my mouth. I put down my cup and left.

Emotion: humiliated

Cluster: Embarrassment
PC1 (Valence): -1.79 Negative
PC2 (Disposition): 0.93

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.370
l0.329
尴尬0.322
ness0.303
because0.298
Suppressed:
de-0.369
own-0.254
😍-0.253
🤩-0.245
cautiously-0.228