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 photograph, innocently displayed on the kitchen counter, seemed to mock her. Her son, smiling, holding a picture of his new teacher: Mr. Thompson. The same Mr. Thompson who’d given her a reputation for being argumentative. The same man who’d made her life, at least at school, a living hell.

She felt a phantom ache in her side, where he’d criticized her posture. She imagined her son going through the same thing. The thought of him in Mr. Thompson's classroom was like a physical blow. She had a sudden, insistent need to protect him, to somehow make him untouchable.

She started writing an email to the school, a request for a classroom change she knew was unreasonable.

Emotion: disturbed

Cluster: Fear / Anxiety
PC1 (Valence): -2.97 Negative
PC2 (Disposition): -0.60

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.318
ness0.309
无法0.300
😞0.297
worse0.284
Suppressed:
de-0.829
la-0.666
a-0.392
B-0.301
"-0.290