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 gallery owner, Mr. Abernathy, had his hand on Elias’s shoulder, a gesture of forced calm. "Just... breathe, Elias. Let's not jump to conclusions." But Elias's breath hitched in his throat, a ragged sound that echoed the tremor in his hands. He stared, utterly consumed by the image. *His* painting, “Crimson Bloom,” replicated perfectly, down to the faintest brushstroke, etched into the smooth skin of a stranger's bicep.

He felt a sudden, dizzying lurch in his stomach. The room seemed to tilt. The vibrant reds and oranges of the tattoo felt like a personal violation. He wanted to scream.

"Who... who is he?" Elias stammered, his voice thin and reedy. Mr. Abernathy said something about a client, a wealthy collector. Elias couldn't process it. His mind was a jumble of anxieties. He could feel heat rising in his chest.

Emotion: panicked

Cluster: Fear / Anxiety
PC1 (Valence): -2.44 Negative
PC2 (Disposition): -0.00

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
worse0.317
😫0.294
😖0.285
😰0.285
escape0.283
Suppressed:
la-0.432
a-0.374
de-0.296
楽し-0.262
happy-0.258