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 dusty attic air hung heavy with the scent of forgotten things. Mark found the list tucked inside an old leather-bound journal of his girlfriend, Sarah’s, grandmother. “Learn to speak fluent Italian.” “Become a published author.” “Fall in love.” The last entry, the one that made his stomach clench, read: “Leave this small town.”

He felt the blood drain from his face, leaving him cold. He ran his hand over the list, the paper thin and fragile. They'd built a life here, in this small town. A life together.

He slowly backed away, needing to escape the confined space, the weight of the attic's silence. He went down the stairs, out into the sunshine. It was all a lot to take in.

Emotion: shocked

Cluster: Surprise / Confusion
PC1 (Valence): -2.19 Negative
PC2 (Disposition): -0.65

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
будто0.298
😰0.268
sickening0.266
не0.266
que0.257
Suppressed:
enthusi-0.283
开心-0.269
ولة-0.262
ارك-0.257
উৎসাহিত-0.255