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 streamed through the window, illuminating dust motes dancing in the air. John, his fingers stained with ink, surveyed his finished painting. It depicted a whimsical forest scene, a swirling symphony of greens and blues. He felt a lightness in his chest, an urge to burst into song. He’d poured his heart into it, and he was exceptionally pleased with the result.

A knock at the door, and his sister, Sarah, entered, holding a drawing. β€œLook what Lily drew today, John!” she announced, her voice filled with amusement. John braced himself.

There, in vibrant crayon, was a familiar figure, standing beside a grinning Lily. A tall, slightly clumsy figure with bright orange hair and a crooked smile. It was Clementine. Clementine, his childhood confidante. A smile spread across his face, so wide it almost hurt. The forest scene he had just painted felt even more vibrant, and he couldn't help but chuckle.

Emotion: happy

Cluster: Positive / Joy
PC1 (Valence): 3.98 Positive
PC2 (Disposition): -1.14

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.676
excitedly0.533
enthusi0.505
de0.498
🀩0.495
Suppressed:
S-0.896
😞-0.430
ness-0.410
γγ‚Œγ§γ‚‚-0.387
L-0.379