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 world felt muted. Each footstep on the freshly vacuumed floor was a weighty effort. The aroma of freshly brewed coffee, usually a comfort, failed to rouse any feelings in her. Amelia shuffled into the kitchen, catching sight of the half-finished novel on the kitchen table. Its cover, a vibrant splash of color, offered a stark contrast to her own internal landscape.

She should have felt anger, perhaps betrayal, but mostly, a profound sense of numbness settled over her. Her husband, George, had been so focused lately, his eyes sparkling with a creative intensity she hadn't seen in years. Now, she understood where it all was going. He’d been working on a book. A book about her.

She touched the cover, her fingers tracing the title. Then she moved on, and poured herself a glass of water, and went to sit by the open window.

Emotion: tired

Cluster: Fatigue / Lethargy
PC1 (Valence): -0.92 Negative
PC2 (Disposition): -1.00

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
weary0.254
fatigue0.249
😴0.238
sighed0.237
barely0.230
Suppressed:
C-0.268
because-0.225
B-0.219
🤟-0.217
భి-0.214