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).

Sunlight dappled the overgrown backyard. Amelia knelt, dusting off the rusty metal box with the sleeve of her t-shirt. Inside, nestled amongst yellowed newspaper clippings and a chipped ceramic mug, was a worn leather-bound journal. She ran a finger over the embossed initials on the cover, a wave of warmth washing over her. It had been decades since she'd penned those letters. A faint smile touched her lips as she thumbed through the brittle pages, finding a dried flower pressed between the entries. A giggle escaped her as she remembered the secret she'd been harboring, the shy girl who'd written this. She felt an unexpected lightness in her chest, a silent thank you to her younger self for the hidden treasure.

Emotion: grateful

Cluster: Positive / Joy
PC1 (Valence): 3.77 Positive
PC2 (Disposition): -0.43

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.571
de0.562
B0.546
மகி0.517
happy0.493
Suppressed:
[]-0.582
😖-0.519
worse-0.509
😫-0.498
😣-0.494