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 coffee tasted like ash in his mouth. Michael hadn’t felt truly awake in weeks. He spent his days in a fog, the world appearing to him in shades of gray. He'd been aimlessly meandering through a flea market, the cacophony of voices and smells usually overwhelming, but today, they were just another layer of the general unease that clung to him. He gravitated towards a table piled high with old books.

A particularly battered edition of Dickens caught his eye. As he flipped through the pages, a small, carefully drawn sketch of a bird fluttered out, followed by a letter. The paper was thin, almost translucent, and the script was tiny, almost microscopic. "Sometimes, the smallest things… the way the light catches… are the only reasons…" the letter whispered, and for a moment, the world didn't feel so heavy.

Emotion: listless

Cluster: Fatigue / Lethargy
PC1 (Valence): -1.03 Negative
PC2 (Disposition): -1.07

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
’0.325
давно0.280
unremarkable0.270
lately0.266
either0.266
Suppressed:
B-0.460
P-0.331
l-0.312
C-0.292
👏-0.273