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 teenager, Maya, leaned against the ancient oak tree, hands stuffed in her pockets. She’d stumbled upon the metal cylinder while hiking, half-buried in the roots. The inscription, barely legible, said “To Myself – 2012.” A wave of confusion washed over her. 2012? She was barely out of elementary school then.

She unscrewed the lid, the metal protesting loudly. Inside, a faded Polaroid photograph rested on top of a stack of letters. She picked up the photo. It was her, younger, smiling widely, holding a stuffed rabbit. A strange lightness seemed to fill her chest. She couldn't tear her eyes away from the image. It felt like looking into a lost mirror.

Emotion: awestruck

Cluster: Surprise / Confusion
PC1 (Valence): -0.17 Negative
PC2 (Disposition): -1.20

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
L0.271
suddenly0.260
stranger0.252
plötzlich0.246
wonder0.242
Suppressed:
la-0.450
싶-0.350
K-0.319
rest-0.248
싶은-0.238