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 spreadsheet lay open on the laptop, the glow illuminating Sarah’s face in the darkened room. Her husband, David, was asleep upstairs. She knew she shouldn’t be doing this. Curiosity, a nasty little gremlin, had gotten the better of her. The numbers swam before her eyes: thousands and thousands of dollars, tucked away in an account she didn’t know existed. A coldness spread through her chest.

She closed the laptop, the click echoing in the silence. It felt like a betrayal, even though no vows had been broken, in the traditional sense. She imagined David, meticulously squirreling away his money, and a knot formed in her stomach.

She slipped out of bed, the floorboards groaning under her weight. She made her way to the kitchen, seeking solace in a glass of water. The tap dripped, a metronomic beat that kept time with the frantic pulse in her ears.

Emotion: guilty

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 1.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.492
L0.398
탓0.354
been0.323
ashamed0.319
Suppressed:
de-0.400
(!)-0.279
la-0.276
😍-0.274
"!-0.264