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).

"Could you please keep it down?" Mark snapped, his voice barely a whisper, as he stepped out onto his porch. The worker barely glanced up from his work. He just shrugged and hammered on. Mark’s palms were slick with sweat. He felt a tremor in his hands.

He retreated inside, his jaw clenching. He hadn't slept well in weeks, and the added clamor was making it worse. He felt a constant pressure behind his eyes, a dull ache that seemed to be expanding. The air in his house felt thick, suffocating. He wanted to scream.

He poured himself a glass of water, his hand shaking so badly he almost dropped the glass. He knew he was overreacting, but he couldn’t seem to help himself. Every loud noise felt like a personal assault.

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