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 scent of crayons and construction paper filled the air, a scent that usually soothed Sarah. Today, it was cloying. Her niece, Maya, was proudly displaying a picture of a stick figure with wings and a crown. β€œIt’s King Floof!” Maya declared, her eyes shining with excitement.

Sarah forced a smile. She found herself picking at a loose thread on her sweater. King Floof. He was supposed to be gone, a relic of her childhood, a friend forgotten in the pursuit of adulthood. But here he was, staring back at her in vivid crayon strokes.

She felt a sudden urge to get away, to be alone. She mumbled something about a headache and promised to play later, backing away from the scene as if the drawing were a live threat. The color drained from her face.

Emotion: ashamed

Cluster: Shame / Guilt
PC1 (Valence): -1.97 Negative
PC2 (Disposition): 1.06

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.376
ashamed0.344
ness0.327
逃0.297
😞0.295
Suppressed:
de-0.539
que-0.300
😍-0.288
(!)-0.269
🀩-0.242