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 attic air hung thick with dust and the scent of forgotten summers. Maya, however, barely noticed. A gasp escaped her lips as she lifted the lid of the ancient trunk. Inside, nestled amongst moth-eaten blankets, were her grade school art projects. Her childish doodles, the construction paper turkeys, the popsicle stick castles – all meticulously preserved.

A giggle bubbled up from her chest. She couldn't help herself. Pulling out a crayon-scrawled drawing of her family, she ran downstairs, calling out, "Mom! Look what I found!"

Her mother, initially surprised, quickly softened into a smile. "Oh, those! I couldn't bear to throw any of them away," she said, her voice laced with fond memories. Maya squeezed her mom in a tight embrace, feeling an expansive joy bloom inside her.

Emotion: thrilled

Cluster: Positive / Joy
PC1 (Valence): 3.31 Positive
PC2 (Disposition): 0.07

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.613
!0.544
B0.508
🤩0.499
🎉0.457
Suppressed:
S-0.601
😞-0.536
ness-0.462
😥-0.351
😔-0.341