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 crisp white envelope felt heavy in Amelia's hand. It had been years, a lifetime even, since the event. Her breath hitched. The address, in a familiar, looping script, was etched in her memory like a brand. She hadn't heard from Sarah in… well, a truly long time. A chill snaked up her spine, despite the late summer heat. She knew, before she opened it, that this was not going to be a simple letter.

Her fingers trembled as she slid the thin paper from the envelope. The words swam before her eyes. Guilt, a cold, cloying taste, flooded her mouth. She dropped onto the nearest chair, the flimsy wood creaking under her sudden weight. She reread the first line. Each word was a tiny hammer blow against her defenses.

Emotion: worried

Cluster: Fear / Anxiety
PC1 (Valence): -2.02 Negative
PC2 (Disposition): -0.43

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
worse0.285
S0.279
😰0.277
😣0.269
Worse0.250
Suppressed:
la-0.476
de-0.432
πŸŽ‰-0.233
happy-0.231
"-0.228