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 dusty bell above the pawn shop door chimed, announcing her arrival. Sunlight, thick with the afternoon, sliced across the cluttered counter, illuminating a glint of gold within a glass display. He was a gruff-looking man, wiping down a shotgun with a practiced hand. She cleared her throat, then pointed at the ring. It was a simple band, a single, slightly cloudy diamond.

“How much for that?” she asked, her voice soft, not wanting to startle him.

He named a price – a pittance. She barely blinked. Her heart ached for the woman who’d lost it, for the memory it represented.

Without bargaining, she counted out the bills. As he wrapped the ring in a small velvet box, she smiled, a genuine, open smile. “Thank you,” she said, her voice filled with a warmth that surprised even herself. “Have a good day.”

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