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 smell of freshly cut grass drifted in through the open window, a sharp contrast to the stale air of the living room. Sarah, my neighbor, was on the porch, her face flushed with activity. She was carrying a clipboard and a pen. "Hi, David," she chirped. "That old maple... it's really starting to overhang."

My limbs felt like they were filled with sand. I gestured weakly towards the armchair, as if the movement itself were a tremendous undertaking. "Yeah, yeah," I mumbled, my voice rough. The idea of going outside, of surveying the tree, of discussing it... it felt like scaling a mountain.

The sun beat down, and the mere thought of being exposed to it made me want to retreat further into the shadows. I watched her, a hive of energy, as she started to explain what she wanted to do. My eyes drifted closed, and I wished, with every fiber of my being, that she would just go away.

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