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 fluorescent lights of the accounting department seemed to hum a mocking tune. Amelia slammed her stapler onto the desk, the metal clack echoing the agitation churning inside her. It had only been a week since the transfer to marketing. A week. And already, the forced smiles, the endless meetings about “synergy,” the bright, airy office space felt like a gilded cage.

She rubbed her temples, trying to massage away the tension that had settled there like a permanent knot. The smell of freshly brewed coffee, usually a comfort, now felt cloying, a sugary sweetness that coated her tongue in a film of resentment. “So, are we ready to brainstorm?” Brenda asked with unsettling cheer. Amelia just stared out the window. The cityscape was a blur of grey.

Emotion: brooding

Cluster: Sadness / Despair
PC1 (Valence): -2.46 Negative
PC2 (Disposition): -0.60

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.749
😞0.329
그의0.299
un0.273
但他0.270
Suppressed:
de-0.917
la-0.603
a-0.369
l-0.335
her-0.330