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).

John stretched languidly on the sofa, the afternoon light painting stripes across his legs. The kids were at their grandparents' house, and Sarah was at a conference. He had the whole house to himself. A feeling of utter contentment settled over him, like a warm blanket. He flipped through channels, eventually landing on a documentary about classic cars. His hand drifted towards the small, locked box he kept on the bookshelf.

He’d always thought it was empty, a relic of some forgotten project. The key, kept in a drawer, felt heavy in his hand as he unlocked the box. Curiosity sparked, something he rarely experienced anymore. Inside, beneath a layer of dust, he found a passbook. The name on the account was Sarah’s. The numbers were… significant.

Emotion: relaxed

Cluster: Calm / Serenity
PC1 (Valence): 2.55 Positive
PC2 (Disposition): -2.48

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.638
disfrut0.418
прият0.386
楽し0.368
affectionately0.368
Suppressed:
L-0.598
S-0.558
C-0.435
because-0.428
ness-0.379