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 day had been perfect. The garden was thriving. The flowers were in full bloom. A light breeze was carrying the scent of the lilac bushes. She’d spent the afternoon watering plants, humming a happy tune. Returning to the house, she found Lily, her daughter, had left her tablet on the couch. A notification popped on the screen. She felt a tickle of curiosity and picked it up.

As the app loaded, a slow chill began to creep through her. The profile was set to private, but the profile picture was unmistakable, a blurry selfie of Lily with her friends. An unfamiliar username. A hidden world.

Her world tilted as she scrolled through the photos. Pictures of vaping, of parties, of a side of Lily she didn't know existed. All of the warmth that she felt was replaced with a sudden and intense coldness. Her hands trembled.

Emotion: vibrant

Cluster: Positive / Joy
PC1 (Valence): 2.83 Positive
PC2 (Disposition): -1.36

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.518
!0.459
!".0.436
de0.434
!).0.409
Suppressed:
S-0.890
L-0.478
P-0.407
H-0.365
😞-0.306