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 chandelier glittered above Clara, casting dancing lights across the room. She felt a kind of restless energy, a need to move, to be in motion. The soft murmur of conversations and the clinking of glasses created a symphony of sound that felt alive. She felt ready to burst with something she couldn't quite name.

She stepped up to the podium, her hands finding a resting place on the shiny surface. She didn't bother to clear her throat. “Mom and Dad, this is a beautiful thing,” she began, her voice already enthusiastic. "To reach this milestone, and to have touched so many lives is something to applaud!"

She recounted a funny story about her father, complete with animated gestures and exaggerated expressions. She finished, laughing herself, and was met with a chorus of appreciative laughter. She couldn't help but feel a deep sense of contentment, a kind of bubbling joy at the heart of the moment.

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