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 invitation sat on the kitchen counter, crisp and elegant. A garden party. Given by the Millers, the couple who lived in the mansion at the end of the street. I felt my lip curl at the sight of it. They were always so *polished*, so *perfect*, so… unreachable.

They’d never acknowledged me beyond a polite nod. I, on the other hand, had watched them, observed their every move, from the new cars to the lavish Christmas decorations. The smugness of it all. I felt a knot in my stomach.

I crumpled the invitation in my hand. I felt a surge of triumph as the paper crackled. They could keep their garden party. I wouldn’t be seen dead in their company. I’d spend the day in my own garden, drinking tea from a chipped mug, and reading a book.

Emotion: stubborn

Cluster: Defiance / Spite
PC1 (Valence): 0.80 Positive
PC2 (Disposition): 3.73

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.780
l0.685
la0.605
either0.580
a0.501
Suppressed:
难以-0.435
终于-0.419
顿时-0.405
竟然-0.396
wondering-0.391