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).

"Another day, another meeting," Emily sighed internally. The words hung in the air, a silent lament. The promotion felt like a distant peak. She felt like someone had wrapped a tight band around her chest, and she couldn't take a deep breath. She glanced across the office to see that her competitor, Ben, was hard at work.

She stared blankly at her computer screen, the cursor blinking like a mocking eye. The project she had to submit was a mountain. She took a deep breath, trying to summon the energy needed to begin the climb.

Her desk, cluttered with paperwork, seemed to mirror the chaos in her mind. She pushed a stack of papers aside, revealing a half-eaten sandwich. It was still there from yesterday. She wasn’t hungry.

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
因-0.273
কারণ-0.272