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 humid air hung heavy in the coffee shop, mirroring the heaviness in Leo’s chest. He'd always looked up to Ben, the supposed head of a successful advertising agency. Now, he'd seen Ben's real life: a series of dead-end projects and mounting debt. He felt his pulse quicken, his head beginning to throb.

He realized now that Ben’s life, which he had envied, was built on a foundation of deception. The world, which he'd always seen as full of opportunities, began to shrink, to confine. He found himself unable to focus, unable to formulate a coherent thought.

He felt the sudden, desperate urge to flee, to escape this suffocating conversation, to leave the falseness behind. He sat rigidly, unmoving.

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