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).

"I made you *kimchi jjigae*," Kevin announced, beaming. It was her favorite, a dish her mother used to make. The aroma, usually a comfort, now felt… suffocating. He’d spent weeks in the kitchen, meticulously copying her mother’s recipe from her old, handwritten cookbook.

He was proud. He kept looking at her for a reaction.

He had learned to make it from her own family's cookbook. She remembered her mother’s gentle scolding when she would be inattentive in the kitchen, but she could not replicate the taste. A wave of a strange, unwelcome feeling rose in her chest. Everything had changed in their relationship. She’d felt safe with him for years, but now… She stood up abruptly. "I… I need some air," she mumbled, fleeing the apartment.

Emotion: sensitive

Cluster: Fear / Anxiety
PC1 (Valence): -2.10 Negative
PC2 (Disposition): -0.75

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
ness0.304
😞0.282
😥0.272
😣0.271
own0.264
Suppressed:
la-0.679
de-0.660
a-0.281
"-0.280
(!)-0.242