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 champagne flute felt heavy in Elias’s hand, the bubbles doing little to lighten the mood. He’d been maneuvering through the crowded reception, a practiced dance of smiles and nods, when *she* appeared. Across the room, bathed in the soft glow of the fairy lights, her familiar face, a ghost from his past, materialized. He felt a constriction in his chest, his breath shallow. He wanted to turn and flee, to dissolve into the floral arrangements, but a smile was already forming on his face, the muscles in his jaw aching.

"Elias! Fancy seeing you here." Her voice, a melodic echo of laughter and shared secrets, sliced through the celebratory din. He managed a weak chuckle, the sound catching in his throat.

"Sarah," he said, the name feeling foreign on his tongue. He saw himself reflected in her eyes, a stranger in a well-tailored suit, a prisoner in his own skin. He took a long, slow sip of champagne, the cool liquid doing nothing to soothe the sudden heat rising in his neck.

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