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).

“Hey, Sarah?” The voice, loud and clear, pierced the quiet of the evening. It was Mark, her other neighbor, standing at the edge of her property. He was asking if he could create a small bee-friendly garden on the side of her house.

Sarah stepped back from the open doorway. The sun, setting behind him, cast a long, looming shadow. She wanted to retreat, to vanish.

She rubbed her arms, suddenly chilled. She didn’t know how to say no, not without causing offense, but the thought of more people, more interaction, was terrifying. She looked at the ground, then the sky, then at the dirt that Mark was pointing towards. “I… I need to check with the HOA,” she finally whispered, her voice barely there.

Emotion: vulnerable

Cluster: Fear / Anxiety
PC1 (Valence): -2.07 Negative
PC2 (Disposition): -0.69

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.327
ness0.319
😞0.299
😔0.277
😣0.258
Suppressed:
de-0.567
la-0.488
l-0.271
(!)-0.267
(!)-0.257