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 ballroom seemed to shrink, the air growing thick and heavy. David gripped the back of a chair, his knuckles white. He hadn't expected to see her here. He'd thought, with time, he'd be able to handle it. He was clearly wrong. Across the dance floor, she laughed, a bright, familiar sound that felt like a punch to the gut.

He wanted to scream, to run. Every surface felt sharp, every touch foreign. The music, which had been a joyful backdrop, now pounded in his ears, a relentless drumbeat marking time. He saw her glance his way, a flicker of recognition, a ghost of a smile.

He forced himself to take a breath, to unclench his jaw. He told himself it was just a party, a moment. But the feeling of being bound to this spot, to this moment, tightened with every passing second.

Emotion: trapped

Cluster: Sadness / Despair
PC1 (Valence): -3.02 Negative
PC2 (Disposition): -0.80

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.573
ness0.401
无法0.366
hopeless0.360
😞0.353
Suppressed:
la-0.740
de-0.376
l-0.339
several-0.323
excitedly-0.322