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 aroma of freshly baked bread filled Daniel’s small apartment. He carefully arranged a plate of pastries, ready for his friend’s visit. His stomach felt light, giddy, with the promise of laughter and good company. He had a new book to discuss, a new recipe to try, and the whole world seemed vibrant and full of possibility.

A knock at the door startled him. The postman held a small, rather official-looking envelope. Curious, he took it. He recognized the sender's handwriting immediately. His father.

The letter inside was an apology, decades in the making. For a childhood mistake, a harsh word, a lack of understanding. A wave of surprising emotion washed over him. The pastries suddenly lost their luster. He had long ago forgiven his father, or so he thought.

Emotion: cheerful

Cluster: Positive / Joy
PC1 (Valence): 4.13 Positive
PC2 (Disposition): -1.19

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.962
de0.597
!0.590
excitedly0.589
!".0.568
Suppressed:
S-1.213
L-0.693
ness-0.492
😞-0.425
বারবার-0.422