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).

David leaned against the peeling paint of his porch, arms crossed, watching Mr. Abernathy’s driver unload groceries from a gleaming SUV. He’d known Mr. Abernathy for years, a retired accountant who always wore the same tweed jacket. David always thought he was just a regular guy. Now, David was staring at the evidence of an elaborate, expensive life.

He felt a tightness in his jaw. The smell of freshly cut grass, usually a comfort, now felt like an irritant. It had become a sign of a life he couldn’t have. He thought about his own meager savings.

He kicked a loose pebble across the yard, the small sound loud in the sudden silence. He went inside and slammed the door, the noise echoing through the empty rooms.

Emotion: restless

Cluster: Fatigue / Lethargy
PC1 (Valence): -0.51 Negative
PC2 (Disposition): -0.25

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
own0.485
’0.363
S0.333
tìm0.316
destination0.272
Suppressed:
B-0.421
заслу-0.297
টুকু-0.211
arada-0.204
background-0.202