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 stale air grew heavy. Michael, normally boisterous, was now silent, slumped against the back wall, his tie loosened. Sarah watched him, a knot forming in her throat. She understood his discomfort. He was always immaculately put together, a man of precise habits. This claustrophobic, unpredictable situation must be like a personal hell.

She recalled his earlier boast about a date that evening. It was a big deal for him. Now, that was likely off the table. A sudden wave of… sadness, a kind of shared disappointment, washed over her.

“Hey,” she said softly, “Don’t worry about your date, they can’t exactly hold this against you.” Sarah found herself unable to crack a joke; the moment didn't feel right. Instead, she added, “It'll be okay.” She then reached into her bag, pulled out a pack of gum, and held it out to him. “Want some?”

Emotion: empathetic

Cluster: Compassion / Love
PC1 (Valence): 1.55 Positive
PC2 (Disposition): 1.05

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
H0.622
B0.603
L0.529
la0.414
sympat0.393
Suppressed:
iostream-0.441
chẳng-0.393
仿佛-0.359
не-0.352
দুপুর-0.344