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 room seemed to spin. Margaret sat across from her best friend, Susan, who had just explained she was joining a Wiccan coven. Margaret's breath hitched, her heart hammering against her ribs. The coffee in her cup sloshed precariously.

She felt a compelling urge to laugh, a hysterical, disbelieving sound that threatened to bubble to the surface. She clamped her jaw shut, willing the feeling to subside. She stared at Susan, searching her face for a hint of a joke, a trick, anything to dispel the growing feeling of unease.

“But…the rituals…the… magic…” she managed to stutter, the words catching in her throat. She gripped the table, knuckles white.

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