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 was stifling, the air thick with dust and the ghosts of forgotten summers. Eleanor wrinkled her nose, but her jaw remained set. She’d known this would be a task, a challenge, but she was more than up to it.

She’d brought a notebook, a pen, and a large bottle of water. Eleanor was prepared. She began to rummage through the boxes, her fingers moving with a speed that belied her petite frame.

She found a collection of vintage dresses. One in particular caught her eye, a silk gown the color of faded roses. She held it up, admiring the delicate lace and intricate beadwork. She imagined the previous owner, wearing it at a dance. It was almost comical.

Emotion: smug

Cluster: Positive / Joy
PC1 (Valence): 3.43 Positive
PC2 (Disposition): 1.71

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de1.109
l0.615
la0.585
😎0.570
πŸ˜‰0.564
Suppressed:
S-0.579
own-0.486
πŸ’”-0.483
😞-0.464
😣-0.455