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).

Sunlight streamed into the attic, illuminating dust motes dancing in the air. Amelia hummed a quiet tune as she carefully sorted through boxes of her mother’s belongings. The air smelled of old paper and dried lavender. A particular trunk, heavy with age, caught her eye. Inside, meticulously organized in plastic sleeves, were all her old school assignments. A genuine smile bloomed across her face as she lifted a primary school drawing of a lopsided house. She gently traced the crayon lines with a fingertip.

Her mother's devotion, expressed in this quiet preservation, warmed her to the core. She found an old book report on "Charlotte's Web," stained with what looked like a tea ring. A wave of tenderness washed over her. She imagined her mother, years ago, carefully saving this, a small piece of her daughter's world.

Amelia gathered the assignments, pulling them close to her chest. She decided she would make a special album, a tribute to her mother's care. It was the least she could do, a small repayment for all the silent gestures of affection that filled her childhood.

Emotion: loving

Cluster: Compassion / Love
PC1 (Valence): 3.72 Positive
PC2 (Disposition): -0.41

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.604
affectionately0.532
함께0.504
🥰0.459
loving0.440
Suppressed:
не-0.484
S-0.458
worse-0.457
[]-0.438
😖-0.417