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 air crackled around Leo as he examined the photograph. He’d been exploring the attic of the old family home, and the find had him practically levitating. The family narrative painted a picture of his grandfather as a stern, unyielding businessman, a pillar of the community.

The photograph showed a very different man. He was in his youth, laughing, arms around a woman with fire-red hair, both of them drenched from a downpour. Behind them, a sign proclaimed a speakeasy.

He could practically feel the warmth of the moment radiating from the image. It was a secret world he'd never imagined existed. The carefully constructed façade of his family’s history was crumbling, and Leo was delighted.

Emotion: excited

Cluster: Positive / Joy
PC1 (Valence): 3.21 Positive
PC2 (Disposition): -0.05

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.809
!0.564
excitedly0.500
🤩0.491
!"0.466
Suppressed:
S-0.721
ness-0.519
😞-0.424
grieve-0.310
😔-0.300