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 waiting room felt far too warm. The air conditioning hummed, a futile effort against the rising tide of heat in her chest. She kept adjusting the strap of her bag, the leather digging into her shoulder. It was a nervous habit, she knew, but she couldn't seem to stop. Sunlight streamed through the blinds, painting stripes across the beige carpet. Each time the door opened, her pulse would quicken, a hummingbird trapped in her ribs.

Her name was finally called. She stood, smoothing down her skirt, and took a deep breath, steeling herself. The office was decorated in muted tones, a stark contrast to the bright, sterile waiting room. And then she saw him. Across the room, behind a large desk, sat Mark. The Mark. Her stomach clenched.

โ€œPlease, come in and have a seat,โ€ he said, his voice measured and calm. A polite smile played on his lips, but it didn't reach his eyes.

Emotion: uneasy

Cluster: Fear / Anxiety
PC1 (Valence): -1.80 Negative
PC2 (Disposition): -0.43

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
๐Ÿ˜ฐ0.240
๐Ÿ˜–0.231
๐Ÿ˜ฃ0.231
๋А๊ปด0.226
worse0.222
Suppressed:
de-0.474
la-0.465
happy-0.261
a-0.255
ecstatic-0.228