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 sunlight streaming through the attic window felt especially warm on Clara's skin. She could barely contain the grin pulling at her lips. The family lore painted Great-Aunt Mildred as a lonely spinster, consumed by her knitting and her cats. But here, in a glossy black and white photo, Mildred was the center of attention.

She was laughing, her head thrown back, surrounded by a group of fashionable young men. The photo showed her in a stylish dress, and Clara felt a thrill course through her.

The air around her seemed electric. What a story this was! What a revelation! Clara had already decided that she needed to get these pictures scanned.

Emotion: delighted

Cluster: Positive / Joy
PC1 (Valence): 3.80 Positive
PC2 (Disposition): -0.26

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
!0.611
🤩0.536
de0.530
ecstatic0.527
🎉0.524
Suppressed:
S-0.687
😞-0.562
ness-0.471
😔-0.385
😥-0.379