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 shovel struck something hard. Arthur grunted, wiping sweat from his brow. Sun blazing down, he’d been hoping for a lost garden gnome, not whatever this was. He’d barely remembered planting anything in this corner of his yard, but the faint memory of a picnic, a laughter, a summer so long ago, sparked a tiny thrill. He knelt, fingers tracing the outline of the metal box. Pulling it free, he practically skipped towards the porch, the heavy object feeling light as a feather. Inside? Oh, the joy! A perfectly preserved baseball glove, identical to the one he’d adored as a boy, a photo of his best friend grinning, and a note written in the messy scrawl he recognized instantly as his own. He was overwhelmed with a warmth spreading through him.

Emotion: ecstatic

Cluster: Positive / Joy
PC1 (Valence): 3.16 Positive
PC2 (Disposition): -0.15

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
!0.588
ecstatic0.560
de0.559
🤩0.548
overjoyed0.547
Suppressed:
S-0.678
😞-0.500
ness-0.405
😔-0.368
😥-0.359