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).

Rain lashed against the attic window, mirroring the storm inside Amelia. She clutched the dusty, leather-bound book, her mother’s handwriting a ghost against the page. "Climb Mount Fuji." That was the first item. Amelia crumpled the paper in her fist. Her mother had never even mentioned wanting to go to Japan. A sharp ache throbbed in her chest. She’d thought she knew her mother.

The silence of the attic was a suffocating blanket. Sunlight, when it finally broke through the clouds, illuminated the dust motes dancing in the air, mocking her. She ran a hand through her hair, snagging on a knot. Why hadn't she been enough? Why hadn't her mother shared any of this with her?

Amelia slammed the book shut, the sound echoing in the stillness. She felt a profound sense of isolation, a chasm opening between the woman she thought she knew and the vibrant, secret life revealed in faded ink. She felt her throat constrict.

Emotion: hurt

Cluster: Fear / Anxiety
PC1 (Valence): -2.46 Negative
PC2 (Disposition): -0.42

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.412
L0.396
ness0.338
😞0.328
😢0.290
Suppressed:
de-0.643
la-0.402
!-0.259
(!)-0.248
/-0.244