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 scent of cinnamon and sugar hung in the air of the bakery, a scent usually associated with comfort, but today, she found it nauseating. "The cake was delicious, wasn't it?" Emily asked, a soft smile on her face. "The best we ever had!"

The cake. The birthday cake. It had been burnt, a charcoal monstrosity. She remembered crying, embarrassed and humiliated.

Her stomach churned. A headache was forming behind her eyes. Emily's version was so sweet, so positive. It felt like she was trapped in a funhouse mirror, where every reflection was distorted and twisted. She wanted to run, to escape the sickening sweetness of the moment.

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