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).

She picked at the corner of her sticky note, the paper tearing under her restless fingers. The office felt vast, almost echoing, even with the keyboard clicks and phone calls. Sarah leaned back in her chair, the leather squeaking in protest. A flyer for a company-wide pizza party, a blatant attempt to boost morale amidst the sale rumors, lay crumpled in her wastebasket.

A cold draft snaked around her ankles. She shivered, pulling her sweater tighter. The faces in the office, once familiar, seemed distant, their conversations indistinct. She'd stopped eating lunch in the break room.

The announcement of the sale hung heavy in the air. She thought of her team, of the shared jokes and inside references. She’d made a few friends. Her stomach twisted at the thought of the future. The pizza party, she decided, was definitely not for her.

Emotion: lonely

Cluster: Sadness / Despair
PC1 (Valence): -1.32 Negative
PC2 (Disposition): -1.29

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.416
loneliness0.375
孤独0.362
solitude0.351
lonely0.344
Suppressed:
de-0.450
l-0.360
/-0.272
👏-0.262
B-0.241