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 attic air tasted of dust and forgotten things. Clara tugged a dusty sheet away from a small wooden rocking horse. Its paint was chipped, a single eye missing, but she carefully righted it. A pang went through her, picturing a child’s glee.

She continued, sorting through boxes labeled with faded names. "The Millers, 1988," one read. Inside, a collection of mismatched teacups. She imagined Mrs. Miller, perhaps hosting afternoon tea parties. The details hinted at life, a whole world that once existed within these walls.

A worn leather-bound photo album lay open. Smiling faces stared back at her, frozen in time. She gently closed it, feeling a deep welling of consideration for the lives that had passed through this house before her own. The past, she realized, was a fragile thing.

Emotion: sympathetic

Cluster: Compassion / Love
PC1 (Valence): 1.13 Positive
PC2 (Disposition): 0.75

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
H0.521
L0.421
B0.401
la0.338
sympat0.305
Suppressed:
iostream-0.334
简直-0.306
再び-0.305
chẳng-0.296
не-0.296