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).

She skipped down the street, her bright yellow dress practically bouncing with her. The thought of tonight, of the juicy burgers and crispy fries at "Burger Bliss," filled her with a lightheartedness. She even found herself whistling a jaunty tune. She practically sprinted the last block, her steps light.

The neon sign above the door flickered erratically, casting a strange light on the empty tables visible through the window. A crudely handwritten sign taped to the door announced the closure. A wave of bewilderment washed over her as she reached the door. It was the same shock she felt when the ice cream shop she frequented as a child closed. That memory quickly faded away as she considered how she will find a new burger joint.

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
๐Ÿคฉ0.495
Suppressed:
S-0.896
๐Ÿ˜ž-0.430
ness-0.410
ใใ‚Œใงใ‚‚-0.387
L-0.379