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 silence of the library was almost deafening. Michael ran a hand through his hair, the gesture automatic, a nervous tic. Sunlight streamed through the tall windows, illuminating dust motes that danced in the air. He found an old envelope in a book he had been using as a bookmark.

It was a letter to his ex-wife, a rambling, emotional plea. He had written it right after their separation, consumed with regret. Now, years later, it felt like reading someone elseโ€™s words.

He reread the words over and over, his chest tightening. He felt a deep sense of shame, of having laid bare his soul to someone who had already moved on. He slowly folded the letter back into the envelope, a gesture that barely registered.

Emotion: gloomy

Cluster: Sadness / Despair
PC1 (Valence): -1.68 Negative
PC2 (Disposition): -1.37

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
๐Ÿ˜ž0.368
ๆ— ๆณ•0.319
๐Ÿ˜”0.300
ใ„ใคใ‚‚0.270
์“ธ0.255
Suppressed:
la-0.432
l-0.419
de-0.398
B-0.294
!-0.259