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).

"So, the genius author finally finished his masterpiece, eh?" Leo drawled, his voice a low, dry rasp. He held the book, a cheap paperback, between two fingers as if it were contaminated. It was his girlfriend's, and the cover displayed a caricature of him, and not a flattering one.

He traced the image of the character with his thumb. A self-absorbed, manipulative cad. The similarities were, of course, painfully obvious. He could feel a coldness spreading in his gut.

He threw the book onto the coffee table. The pages fanned out, displaying the fictional, and clearly judgmental, account of their life. His jaw clenched. He’d spent years attempting to be a better person, and all the while, she’d been meticulously documenting his flaws. He picked up his keys and stalked out, the front door slamming behind him. He didn't know where he was going, but he knew he couldn't stay.

Emotion: bitter

Cluster: Shame / Guilt
PC1 (Valence): -1.78 Negative
PC2 (Disposition): 0.84

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.447
striving0.238
😞0.236
aggravated0.225
soport0.223
Suppressed:
own-0.316
de-0.291
la-0.231
🤩-0.227
是非常-0.225