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 thought of going to Sarah's apartment filled me with a deep reluctance. Getting ready felt like scaling a mountain. Every movement was an ordeal. I shuffled into my clothes, my movements slow and deliberate. Even the simplest tasks seemed to demand a monumental effort.

I arrived at her apartment, where she was already bustling about, making cocktails. Her energy was infectious, and it always left me feeling drained. She was a powerhouse, always bragging about her work at a tech startup. Her stories, full of innovation and excitement, had long been a source of both fascination and envy for me.

“I just landed a huge partnership!” she announced, her eyes sparkling. I forced a smile. I looked around her apartment. She had always gone on about the luxury of her home, with the incredible view. She showed me the office, and the desk, with a massive new computer. Everything was so perfect. I went to the kitchen to grab a glass of water when I noticed something, a flyer for a local art exhibit, and a picture of a guy I knew from school.

Emotion: sluggish

Cluster: Fatigue / Lethargy
PC1 (Valence): -0.56 Negative
PC2 (Disposition): -1.03

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
sluggish0.263
dreaded0.258
似乎0.257
буд0.256
barely0.254
Suppressed:
P-0.384
B-0.379
భి-0.246
羞-0.243
ຮ-0.238