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 attic air hung thick and dusty, catching in Eleanor’s throat. A single shaft of sun illuminated motes of dust dancing in a forgotten corner. She was supposed to be sorting through old boxes, a task she’d been putting off for weeks. Each item felt like a weight, a reminder of the life she’d lived and the one she was living now, a life that felt…drained.

She coughed, waving a hand in front of her face. A small, bound book lay half-hidden beneath a stack of yellowed photographs. Curiosity, a rare visitor these days, pricked at her. Tugging it free, she recognized her own handwriting, a frantic scrawl she hadn't seen in decades. It was an unsent letter to her mother, written the year her father left.

The words, though faded, leaped off the page. *I can’t seem to get out of bed in the morning*, it read. She remembered the feeling. The blankets felt like lead, the world a gray watercolor painting.

Emotion: depressed

Cluster: Sadness / Despair
PC1 (Valence): -2.06 Negative
PC2 (Disposition): -1.08

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.417
每天0.353
无法0.330
own0.311
😔0.309
Suppressed:
de-0.492
B-0.393
P-0.352
l-0.310
!-0.307