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).

Liam slammed the lid shut, the cardboard box rattling with the force. He'd found it while clearing out his grandmother's house after her death. Inside, nestled amongst other mementos, was a stack of his mother’s childhood writings, essays, and stories. The penmanship was exquisite, the ideas brimming with creativity. Liam felt a strange pressure in his chest.

He remembered his own struggles with writing, the crumpled drafts, the red ink splattered across pages. He’d always felt like a clumsy writer. He pictured his mother, a literary genius, praised by teachers.

"Amazing," he muttered, picking up a story about a brave knight.

Later, he found himself staring at his reflection in the bathroom mirror, scrutinizing his own unremarkable features. He wished he had the same grace.

Emotion: envious

Cluster: Shame / Guilt
PC1 (Valence): -0.65 Negative
PC2 (Disposition): 1.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.550
H0.431
B0.406
L0.403
envy0.365
Suppressed:
再び-0.262
мед-0.258
再次-0.246
own-0.239
พร้อม-0.231