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 rain gently pattered against the windowpane as she sat by the fireplace, the crackling flames casting dancing shadows on the walls. She feltโ€ฆ serene. The world outside, a symphony of grey, seemed distant, unimportant. She was safe, warm, and embraced in the cozy tranquility.

The letter had surprised her. It was from her former boss, a woman she had thought hated her. It was an admission of wrongful behavior that had caused her much distress. The letter, however, was a healing salve to her.

She picked up her cup of tea, savouring the warmth that emanated through the ceramic. Her shoulders were relaxed, and her muscles eased.

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
๐Ÿ˜0.397
๐Ÿฅฐ0.392
overjoyed0.384
Suppressed:
S-0.724
L-0.398
P-0.394
๐Ÿ˜ž-0.385
๐Ÿ˜ฃ-0.360