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 deadline was looming, and the tension in the office was thick enough to cut with a knife. Mr. Harrison, their boss, was particularly stressed. He kept pacing, his face flushed.

Seeing his distress, Sarah quietly started making coffee. She knew he preferred it strong, so she brewed a fresh pot, adding an extra scoop of grounds. When the pot was ready, she placed a steaming mug on his desk, her actions speaking for themselves.

He stopped pacing, surprised. “Thanks, Sarah,” he mumbled, his voice tight. “Really appreciate it.”

She simply gave a small nod and returned to her own work, feeling a curious lightness. Later, during a brief break, Mr. Harrison casually mentioned Camp Willow Creek. She perked up. They had both spent their childhoods at the same camp.

Emotion: kind

Cluster: Compassion / Love
PC1 (Valence): 4.35 Positive
PC2 (Disposition): 1.39

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.741
B0.692
de0.492
affectionately0.485
latter0.475
Suppressed:
此刻-0.550
не-0.521
désormais-0.491
S-0.465
хочется-0.458