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 late afternoon sun cast long shadows across the park as Sarah nervously adjusted her scarf. She’d been chatting with “Nova” for over a year, sharing everything from her favorite books to her deepest anxieties. Now, Nova was meeting her at the fountain. Sarah’s palms were slick.

A figure emerged from behind a cluster of trees, and Sarah's heart hammered against her ribs. She took a step back, instinctively. The person was tall, maybe a little too tall. Their face was obscured by the shadows of a large baseball cap.

“Hey,” a voice called out. Sarah's breath hitched. She should have picked a different spot. She should have just cancelled.

The person began to approach, and Sarah froze, unable to move.

Emotion: afraid

Cluster: Fear / Anxiety
PC1 (Valence): -2.76 Negative
PC2 (Disposition): -0.70

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😣0.316
마다0.314
worse0.313
😰0.303
恐惧0.290
Suppressed:
de-0.691
la-0.562
a-0.402
"-0.316
brag-0.293