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).

He had left it on the kitchen counter, tucked under a cookbook. It was Sarah’s bucket list. David stared at the neat script, each bullet point a small, sharp stab. *Learn to speak Italian.* *Volunteer at an animal shelter.* He felt a tightness in his chest, like a fist was squeezing his heart.

He’d always thought he knew her. They had been together for ten years, yet the list whispered of a life he wasn't a part of. *Write a novel.* He picked up the cookbook, his fingers tracing the title. He hadn’t noticed her writing lately, always engrossed in her phone or staring out the window.

A wave of dizziness washed over him. He sank into a chair, the wooden seat offering little comfort. He thought of their last argument, a stupid squabble about taking out the trash. He hadn't bothered to listen when she tried to explain, her voice quiet and weary.

Emotion: vulnerable

Cluster: Fear / Anxiety
PC1 (Valence): -2.07 Negative
PC2 (Disposition): -0.69

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.327
ness0.319
😞0.299
😔0.277
😣0.258
Suppressed:
de-0.567
la-0.488
l-0.271
(!)-0.267
(!)-0.257