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 key slid smoothly into the lock, a satisfying *click* echoing in the sudden quiet of the hallway. Sarah bounced on the balls of her feet, a wide smile stretching across her face. The scent of cinnamon and something vaguely floral, probably her mother’s potpourri, filled her nostrils. Home. After five years, she was home.

She dropped her bags with a thud, then spun around, arms outstretched, and yelled, “Mom! Dad! I’m here!” A giggle bubbled up from her throat.

Sunlight streamed through the living room window, illuminating dust motes dancing in the air. Her father emerged from the kitchen, wiping his hands on a dishtowel. “Sarah! Welcome home, sweetheart!” He enveloped her in a hug, his familiar scent of aftershave and coffee a comfort.

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