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 late afternoon sun warmed Elara’s face as she strolled across the quad. She paused, watching the leaves of the ancient oak shimmer in the breeze. A cup of chamomile tea steamed gently in her hands. The quiet hum of students’ conversations and the distant chime of the bell tower seemed to blend into a single, pleasant soundscape. She had just finished a productive morning in the lab, a welcome break from the challenging chapter of her thesis.

She felt a soft tug on her arm. Professor Davies, her advisor, stood before her, his usually brisk demeanor softened. He cleared his throat. "Elara, could you come by my office for a moment?" His smile was small, almost shy.

The news landed with a gentle thud, like a fallen leaf. Professor Davies was leaving for a position at a research institute in Switzerland. Her immediate reaction was a quiet, internal acceptance, a feeling akin to the slow, steady rhythm of her own breathing. She took another sip of tea, the warm liquid tracing a comforting path down her throat. "Of course," she said, her voice calm. “When should I come by?”

Emotion: serene

Cluster: Calm / Serenity
PC1 (Valence): 2.59 Positive
PC2 (Disposition): -2.73

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
disfrut0.469
enjoyed0.430
enjoy0.419
serene0.411
楽し0.410
Suppressed:
S-0.611
L-0.502
de-0.478
H-0.455
worse-0.424