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 envelope, addressed in her mother's elegant hand, felt like a lead weight in Sarah's grasp. The paper within, a delicate stationery, hinted at a different era, a different person. As she unfolded the letter, she felt a tightness in her chest, a feeling she often experienced now when she felt overwhelmed.

Her mother's words, filled with youthful aspirations, spoke of artistic ambitions, a desire to break free from the confines of marriage and motherhood. Sarah looked around the immaculate living room, the perfectly arranged furniture. She swallowed hard.

She wanted to scream, to shatter the placid facade she had cultivated for so long. She found herself clutching the letter so hard that her knuckles turned white. A strange heat flushed her face.

Emotion: trapped

Cluster: Sadness / Despair
PC1 (Valence): -3.02 Negative
PC2 (Disposition): -0.80

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.573
ness0.401
无法0.366
hopeless0.360
😞0.353
Suppressed:
la-0.740
de-0.376
l-0.339
several-0.323
excitedly-0.322