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).

“Another box,” Clara muttered, the words raspy. Her voice felt thick, like she’d swallowed sandpaper. The new house, a sprawling mansion, mocked her with its grand scale. After months of endless packing and selling, she felt like a shell.

She dragged a box into the dusty library, collapsing onto the floor. As she leaned against the wall, her gaze fell on a large, intricately carved bookcase. She ran her fingers across the wooden facade, noticing a slight give in one of the panels.

With a gentle push, the panel swung inwards, revealing a hidden space behind the bookshelf. The room was dark and silent, a forgotten sanctuary. Clara just slumped against the wall, the effort of everything hitting her like a wall of water.

Emotion: weary

Cluster: Fatigue / Lethargy
PC1 (Valence): -1.52 Negative
PC2 (Disposition): -1.02

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.384
😞0.288
sighed0.230
ness0.210
😪0.203
Suppressed:
de-0.339
l-0.300
!-0.222
la-0.221
🤟-0.198