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 invitation arrived, sleek black cardstock, embossed with gold lettering. Sarah traced the elegant script: *“An Evening of Decadence, Celebrating the Birthday of Amelia Thorne.”* A wave of… something… washed over her. It was the same day. Amelia’s birthday was the same day as hers. Sarah felt a peculiar lightness in her chest, a sensation akin to the floor dropping out from beneath her.

She blinked, rereading the date. Same. Day. Her meticulously planned pool party, complete with flamingo floaties and a taco truck, suddenly felt… deflated. She sank into her plush armchair, the invitation slipping from her numb fingers.

“Well,” she murmured, her voice barely audible, “that’s… a thing.”

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