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 swam. Sarah ran her hand through her hair, clutching at her scalp as if she could pull the anxiety out through the roots. Her mother’s attic, normally a source of fond memories, now felt like a prison. The boxes were stacked precariously high, threatening to topple over and bury her in a sea of childhood traumas.

Each paper was a time capsule, meticulously sealed and preserved. Her mother's careful penmanship, the endless notes in the margins, the teacher’s red ink critiques—all of it. She couldn’t breathe.

She kicked at a box, sending papers fluttering into the air like startled birds. A photograph, wedged in among the clutter, tumbled out. A picture of her as a toddler, smiling at the camera. The smile was jarring. She couldn't feel it anymore. It was all a lie, she thought.

Emotion: hysterical

Cluster: Fear / Anxiety
PC1 (Valence): -3.19 Negative
PC2 (Disposition): -0.45

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
worse0.401
胀0.401
점점0.396
😫0.377
それでも0.373
Suppressed:
la-0.672
a-0.618
de-0.541
😎-0.370
😎-0.359