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).

Michael watered the petunias in his window box, the rhythmic motion a soothing balm to his thoughts. The vibrant blooms swayed gently in the breeze. He surveyed his lawn, newly mowed and pristine. He’d just finished a new novel, and his mind felt pleasantly blank.

The phone rang, and it was a detective. They wanted to discuss his address. Michael leaned against the doorframe, a picture of domestic tranquility. He agreed to meet them at the house.

The detective’s words, about a shadowy past, a murder, details he hadn't known, seemed distant. He straightened a picture frame. He invited them to sit, offering them refreshments. He listened, absorbing the information like the soft sponge of the couch.

Emotion: calm

Cluster: Calm / Serenity
PC1 (Valence): 2.72 Positive
PC2 (Disposition): -1.82

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
मुस्कुरा0.415
enjoying0.405
leisurely0.388
:)0.382
কিছুক্ষণ0.382
Suppressed:
S-0.527
L-0.499
worse-0.454
even-0.422
😣-0.420