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).

"Another one," George grumbled, staring at the pile of books he was supposed to be cataloging. The smell of dust and decaying paper clung to him like a shroud. He’d been doing this for hours.

He picked up a volume of travel writing, sighing as he looked at the cover. He flipped through the pages, needing a break.

A bookmark, handmade from what looked like pressed flowers, slipped out. A letter was attached. George scowled.

He picked it up, feeling a familiar surge of annoyance. He wanted to throw it away.

He didn't. He looked at the handwriting, and the tiny drawing of a sailing boat. He started to read. He stopped, and read again. He slowly sat down.

Emotion: grumpy

Cluster: Anger / Hostility
PC1 (Valence): -0.65 Negative
PC2 (Disposition): 0.64

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
👎0.232
either0.230
😠0.227
C0.226
:(0.216
Suppressed:
超越-0.206
newfound-0.197
স্বীক-0.195
खोले-0.192
অপহরণকারীদের-0.184