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 aroma of cinnamon and spices filled the air as David stirred the apple pie filling. His laughter echoed through the kitchen as he worked. His home, a cozy haven of warmth and light, always brought him a feeling of happiness. He felt a deep sense of security, of being exactly where he was meant to be.

A woman with a clipboard and a stern gaze arrived at his front door. She explained the history of the house, how it had been the location of a brutal crime. She spoke of blood and tragedy.

David listened to her story, a tiny frown forming on his brow, and then it disappeared. He took another bite of his pie filling, and smiled. He told her a vague story. β€œThat was a long time ago,” he said with a shrug. β€œThe pie is almost ready.” He looked forward to a quiet evening.

Emotion: blissful

Cluster: Positive / Joy
PC1 (Valence): 3.20 Positive
PC2 (Disposition): -1.79

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
joyful0.417
affectionately0.400
😍0.397
πŸ₯°0.392
overjoyed0.384
Suppressed:
S-0.724
L-0.398
P-0.394
😞-0.385
😣-0.360