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).

He stared at the blank screen, the cursor blinking mockingly. The deadline for his novel was fast approaching, and writer's block had become his unwanted companion. He'd tried everything – walks, coffee, staring out the window – but the words wouldn't come. He felt a profound sense of inadequacy, a crushing fear of failure.

He was rummaging through a box of old mementos, hoping for inspiration, when he found it. A letter, addressed to his college professor, detailing his youthful ambitions. He couldn't help but smile; the handwriting and youthful enthusiasm were undeniable.

The letter was bursting with a naive certainty, a belief in his own genius. Reading the words felt like a punch to the gut. The distance between the aspiring writer and the struggling one felt vast. The weight of expectation felt unbearable. He felt a sharp, persistent ache in his temples.

He closed his eyes, the words of the letter swirling in his head. Then he crumpled the paper in his hand, a small act of defiance. He pushed himself away from the desk. Maybe a walk would help. Or maybe not.

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