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 hum of the fan was the only sound in the small, cluttered apartment. Emily sat glued to the couch, a half-finished book resting on her lap. The sheer effort of reading, of thinking, was exhausting. A persistent thumping from downstairs reminded her of the noise the neighbors made. She hated it. She hated everything.

The thumping resolved itself into a knock on her door. It was the building manager. He held a dirty box. "Found this digging under your window. Must have been buried a while."

Inside, she discovered a music box, with a missing key, a collection of old postcards from places she'd never been, and a single, faded photograph. She closed the lid. All too much effort. Back to the couch.

Emotion: lazy

Cluster: Fatigue / Lethargy
PC1 (Valence): -0.37 Negative
PC2 (Disposition): -0.02

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😴0.458
snooze0.425
惰0.401
Lazy0.392
躺0.386
Suppressed:
able-0.371
感激-0.289
because-0.274
å› -0.273
কারণ-0.272