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’d avoided the attic for weeks. The thought of it, a shadowy void of the past, filled him with a quiet dread. Now, compelled by a nagging feeling, he climbed the creaking stairs. He found a stack of old love letters tucked away in a dusty hatbox. They were addressed to a “Dearest Clara,” filled with the fervent adoration he couldn’t recall ever inspiring in anyone.

His shoulders slumped as he read the flowery prose. The attic felt vast and echoing, a stark contrast to the small, cramped apartment he occupied, the walls closing in on him. He felt the familiar weight in his stomach, a dull ache that followed him like a shadow. He carefully placed the letters back in the box, the vibrant words a painful reminder of a life he hadn't lived.

Emotion: worthless

Cluster: Sadness / Despair
PC1 (Valence): -2.10 Negative
PC2 (Disposition): -0.15

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
L0.427
S0.405
😞0.386
ness0.339
缺乏0.326
Suppressed:
de-0.320
(!)-0.286
随后-0.268
😍-0.262
tc-0.255