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).

Sarah found herself staring at the artwork again, the small, crudely drawn shape mirroring the vivid memories she’d long buried. It was the floppy-eared rabbit, Thumper. She hadn’t thought about him in years, but now, seeing him brought back a forgotten unease. Her heart hammered a frantic rhythm against her ribs. She felt a wave of dizziness wash over her.

She grabbed the drawing, the cheap paper crinkling under her fingers. Her niece, Maya, was happily building a tower of blocks, but Sarah couldn’t focus. The image of Thumper seemed to bore into her mind, a nagging question mark she couldn't answer. She felt a sudden urge to get out, to distance herself.

Emotion: unnerved

Cluster: Fear / Anxiety
PC1 (Valence): -2.36 Negative
PC2 (Disposition): -0.55

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😰0.302
😣0.282
будто0.265
😖0.265
ness0.257
Suppressed:
de-0.585
la-0.495
a-0.403
happy-0.289
ecstatic-0.284