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).

He slammed the trunk shut, the metal echoing in the quiet suburban street. The air itself seemed to crackle with his internal turmoil. Seeing that bear, Rufus, nestled amidst a pile of discarded toys, had felt like a punch to the gut. It had belonged to him, a constant companion through a painful childhood. The woman, oblivious, chattered about “vintage finds.” He had tried to be civil, to negotiate, but his voice had grated, each syllable a strained effort. Now, he fumbled with the key, a tremor running through his fingers as he unlocked the door. He’d pay any amount to have that bear back, just to rip its stuffing out.

Emotion: mad

Cluster: Anger / Hostility
PC1 (Valence): -1.94 Negative
PC2 (Disposition): 1.26

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.448
骂0.344
est0.340
🤬0.336
aggravated0.332
Suppressed:
a-0.495
de-0.485
la-0.387
H-0.344
latter-0.290