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 weight of the warm, fresh towels felt good on her hands as she folded them. The aroma of lavender and sea salt filled the room. She surveyed the spa, a sense of deep gratification settled in her. A knock on the front door startled her. It was Mr. Evans, his face etched with worry. "Excuse me, ma'am, but that giant elm… it’s dropping too many leaves in my swimming pool. Do you think we could talk about maybe…?" She put the towels down gently and opened the door. “Mr. Evans, I'd rather not cut down that gorgeous elm if it can be avoided. We can prune the branches overhanging your property to address the leaf issue. Let me call for a professional evaluation and find out what we can do."

Emotion: fulfilled

Cluster: Positive / Joy
PC1 (Valence): 3.22 Positive
PC2 (Disposition): -1.74

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.819
மகி0.415
happy0.397
grateful0.389
de0.385
Suppressed:
S-0.599
worse-0.465
C-0.460
either-0.436
😣-0.426