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).

She stood on the back patio, breathing deeply, the fresh air filling her lungs. The garden she had planted was thriving, bursting with color and life. The new house was a refuge, a haven. She felt… renewed. The world felt brighter, more promising than it had in a long time.

A man from the local historical society knocked on the door. He asked if he could come in to examine the house. He said, with a polite smile, that he was researching a case.

As he began to move around the house, measuring the walls and taking notes, he revealed the information. It turned out the house was tied to a notorious cold case. She watched the man, her sense of well-being draining away with each detail he shared. The garden suddenly looked less inviting.

Emotion: refreshed

Cluster: Positive / Joy
PC1 (Valence): 3.31 Positive
PC2 (Disposition): -1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.998
de0.703
!).0.384
(!)0.381
!".0.379
Suppressed:
S-0.782
C-0.478
L-0.414
😣-0.388
😞-0.376