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 attic air tasted of dust and forgotten things. Clara pulled the tattered cardboard box closer, the contents promising a glimpse into her mother's past she hadn't seen before. Among faded photographs and moth-eaten shawls, a single sheet of paper, neatly folded, caught her eye. It was her mother's handwriting, crisp and familiar, yet the words… a resignation letter. From a job she had always said she loved?

Her breath hitched. She sank onto a nearby trunk, the wind knocked out of her. The letter, dated 1983, detailed grievances she'd never voiced, frustrations she'd never shared. A knot formed in Clara’s stomach.

The accusations, aimed at a superior she’d always spoken of with fond amusement, felt like a thunderclap. Her hand trembled as she scanned the carefully penned paragraphs, piecing together a different version of her mother’s life, a hidden narrative.

Emotion: astonished

Cluster: Surprise / Confusion
PC1 (Valence): -0.84 Negative
PC2 (Disposition): -0.24

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
que0.448
de0.350
la0.323
l0.307
incred0.266
Suppressed:
own-0.289
cheerfully-0.231
enjoyed-0.224
开心-0.221
ี่ยว-0.219