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 morning had started off perfectly. The sun peeked through the blinds, casting a warm glow across Amelia’s face. Coffee was brewing, and the scent of pancakes filled the air. Today was Leo's first day of middle school, and Amelia's mood was absolutely bright.

She leaned in to kiss Leo goodbye, and then heard his teacher's name. Mrs. Gable. *Her* Mrs. Gable, the one who saw her struggle and helped her learn the way she needed.

A smile stretched across Amelia's face, wide and genuine. She felt a profound sense of… well-being. "Leo, you are in the best hands," she said. It was more than a feeling; it was a certainty.

Emotion: joyful

Cluster: Positive / Joy
PC1 (Valence): 3.96 Positive
PC2 (Disposition): -0.67

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.615
de0.600
!0.572
excitedly0.552
enthusi0.533
Suppressed:
S-0.893
ness-0.513
😞-0.470
😣-0.337
L-0.334