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 musty smell of the basement was familiar to Leo. He felt a soft, pleasant weariness as he explored the boxes, relics of his family's history. He’d never felt the need to rush through the task.

He found a set of photos of his father, a man he remembered as reserved and emotionally distant. In the pictures, his father was a wild, carefree young man. Smiling and laughing, a beer in his hand. Leo felt a languid acceptance of this new information.

Then came a photo of his parents, both wearing clothes he didn’t recognize, at a location he couldn’t identify. His mother, the woman who had always seemed to be his rock, was looking at the camera with what Leo recognized as a mischievous glint. It created no urgency in him to ask questions. He put the photos down and let the quietness of the basement surround him.

Emotion: docile

Cluster: Passivity
PC1 (Valence): 0.25 Positive
PC2 (Disposition): -1.34

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
ค่อย0.311
淡淡0.302
enjoy0.273
enjoyed0.273
สบาย0.272
Suppressed:
la-0.589
de-0.413
l-0.373
!!-0.344
!!!-0.316