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 rejection letter, crisp and impersonal, lay on Elara's kitchen table. A single, thin ray of sunlight pierced the blinds, illuminating the dust motes dancing in the air. She poured herself a cup of chamomile tea, its warmth spreading through her hands. The words on the page stung, but she didn’t crumble. She’d spent years on this novel; months revising after the first draft. She took a slow sip, letting the bitterness of the tea linger on her tongue. It was a good exercise in endurance.

She rose, stretching her arms above her head, feeling the satisfying crackle in her spine. The manuscript would go out again. Perhaps with a few more changes. She picked up the phone, her fingers finding the number of a writer's group she belonged to. They could offer perspective, or simply, a sympathetic ear. Elara knew that the next step was writing the next letter.

Emotion: patient

Cluster: Passivity
PC1 (Valence): 2.71 Positive
PC2 (Disposition): 0.40

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
rest0.467
eventually0.415
leisurely0.369
慢慢0.357
a0.355
Suppressed:
无法-0.432
even-0.428
인해-0.421
無法-0.415
😣-0.391