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 gripped the book, the glossy cover slick against her sweaty palms. Her picture, a childhood photo with missing teeth, stared back at her with unsettling familiarity. She flipped through the pages, her breath catching in her throat. The prose, flowery and dramatic, detailed her first kiss, a memory she'd kept locked away, a secret shame. A wave of nausea rolled through her, forcing her to clutch the edge of the kitchen table.

She flung the book across the room. It landed with a soft thud against the wall. This stranger, this *author*, knew things. Things no one should know. She felt exposed, stripped bare. Her hands trembled as she reached for her phone, dialing her sister's number, the words catching in her throat.

Emotion: horrified

Cluster: Fear / Anxiety
PC1 (Valence): -2.63 Negative
PC2 (Disposition): -0.17

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😰0.336
sickening0.324
worse0.314
😣0.307
😖0.296
Suppressed:
la-0.444
B-0.295
enjoyed-0.260
optim-0.246
happy-0.244