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 sun beat down on the freshly tilled soil of the community garden. Sarah, her face hidden beneath a wide-brimmed hat, mechanically weeded the rows of vegetables. It was a task she had taken up after her husband, George, passed away. She found a sort of solace in the repetition. A metal thud echoed as her trowel struck something hard.

It was a small, cylindrical container, unlike anything she'd seen before. She managed to pry it open. Inside, a USB drive and a note. The note read: β€œMemories for a rainy day.” She drove home and plugged the drive into her computer. The images flickered across the screen. George, alive, laughing, and she. The world spun.

Emotion: grief-stricken

Cluster: Sadness / Despair
PC1 (Valence): -2.10 Negative
PC2 (Disposition): -1.14

Role in Research

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

Logit Lens (Layer 40)

Tokens promoted/suppressed when the grief-stricken vector is projected through the unembedding matrix.

Promoted:
😞0.396
ζ³ͺ0.338
😒0.332
haunting0.331
πŸ˜”0.325
Suppressed:
de-0.434
😎-0.330
B-0.303
πŸ€“-0.285
😎-0.276