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 antique music box gleamed under the overhead light, its inlaid mother-of-pearl catching the rays. Uncle George was rambling on, explaining the provenance of the piece, the details of its mechanism. I steepled my fingers, listening patiently, the slight twitch in my lip the only indication of my contentment. He wanted to sell it – to fund his new… venture. Good riddance, I thought, to that trinket.

“So, what do you think, Amelia?” Uncle George finished, beaming. He expected a protest, a plea.

I unfolded my hands and smiled. “It's… lovely, Uncle. And yes, absolutely. You should do what you need to do.” His face fell, he opened his mouth, but I kept my tone breezy. “Here,” I said, pointing at the bill of sale already prepared and on the coffee table. “I’ll fill out the relevant spaces. Just your signature.”

Emotion: satisfied

Cluster: Positive / Joy
PC1 (Valence): 3.20 Positive
PC2 (Disposition): -1.31

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.518
happy0.465
enjoy0.439
relaxed0.430
enjoys0.429
Suppressed:
S-0.515
C-0.511
😣-0.510
worse-0.477
😖-0.453