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 silence of the hospital room was oppressive. Michael sat beside his father's bed, the letter clutched in his sweaty hand. His father, frail and barely conscious, had never mentioned his own father, Michael's grandfather. Until now.

He reread the words, his eyes scanning the lines of faded ink. He felt a strange unease, like a shadow falling over his heart. He wanted to understand, but everything was vague.

He watched his father's chest rise and fall, the rhythmic breaths a stark contrast to his own racing pulse. He felt a rising sense of unease, a chill that had nothing to do with the room's temperature.

Emotion: alarmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.07 Negative
PC2 (Disposition): -0.14

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😰0.304
😨0.272
worse0.269
😣0.261
не0.254
Suppressed:
de-0.323
enthusi-0.258
happy-0.235
delight-0.225
ية-0.217