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 snatched the lab report from the technician’s hand, the paper crinkling in his grip. “This is a mistake,” he stated, his voice tight. It just had to be a mistake. He’d lived a healthy life, never abused his body. He’d followed every rule. He hated injustice, and this was an injustice. The technician, barely out of her teens, blinked at him, her face a mask of polite concern.

He noticed the other patient, a woman with vibrant pink hair, who was already chatting with the doctor. She was laughing! He felt his pulse thrum against his wrists. She looked entirely too relaxed.

The doctor, a young woman with a reassuring smile, looked at him. He couldn’t even bring himself to meet her eyes. He wanted to be anywhere but here, confronting this unbelievable reality. He wasn’t ready to share a club with a stranger, especially a stranger who was probably handling things better than he was.

Emotion: indignant

Cluster: Anger / Hostility
PC1 (Valence): -0.44 Negative
PC2 (Disposition): 2.14

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
l0.614
C0.369
de0.305
que0.295
😠0.288
Suppressed:
own-0.363
soon-0.302
optimistic-0.270
உற்ச-0.261
gradually-0.256