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).

He bounced on the balls of his feet, the eviction notice crumpled in his fist. A grin stretched across Marco's face, wide and unwavering. The cramped, stale air of the apartment seemed to vibrate with a newfound energy, an electric hum that ran through his veins.

He kicked the door shut with a satisfying thud. Finally! The shackles were off. He could practically taste the open road, the freedom to roam. He felt a surge of adrenaline, a desire to run and run until he collapsed from exhaustion.

He tossed the crumpled paper onto the kitchen counter and went to turn on some music, which he blared at a loud volume. His neighbors would hate it, but for the first time in a while, Marco didn’t care.

Emotion: jubilant

Cluster: Positive / Joy
PC1 (Valence): 3.66 Positive
PC2 (Disposition): -0.27

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.738
la0.707
!0.670
ecstatic0.547
excitedly0.539
Suppressed:
S-0.768
😞-0.485
ness-0.429
😔-0.376
😥-0.351