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).

"Just breathe, Sarah," she muttered under her breath, her voice barely audible. The pressure in her chest intensified. She'd been clearing out the garage when she found it: a metal container, sealed tight and covered in rust. There was no recognition whatsoever.

Inside, a collection of carefully folded letters, a dried flower pressed between the pages of a slim volume of poetry, and a key with an ornate design. It took a while to find out what the key was for: her childhood home. She began opening the letters. The handwriting was neat, elegant, yet alien.

The words spoke of a shared childhood, of shared secrets, of a love she couldn’t place. She sat there, paralyzed, the garage suddenly feeling too small, too close. The unfamiliar memories, the emotional ties she couldn't understand – the weight of it threatened to suffocate her. She felt like she might break into a thousand pieces.

Emotion: stressed

Cluster: Fear / Anxiety
PC1 (Valence): -1.37 Negative
PC2 (Disposition): -0.53

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😩0.266
😫0.256
😰0.239
😖0.237
😞0.236
Suppressed:
B-0.200
happy-0.197
enjoy-0.176
enjoying-0.172
bahagia-0.169