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).

He leaned back in the chair, hands clasped loosely in his lap, the polished wood cool against his palms. The blinds were drawn, casting the room in a gentle, hazy light that diffused the harsh edges of his anxiety. He’d meticulously chosen the playlist – ambient music, tailored to his needs.

He closed his eyes, savoring the tranquility that had finally settled over him after weeks of turmoil. It was a private sanctum he’d cultivated, a place where the world’s chaos couldn't reach him. He felt…contained.

A knock. His eyes snapped open. “Come in,” he managed, his voice steady. The door opened and there she was, the girl who'd made his adolescence a living hell. Jessica. He felt a sudden, sharp ache in his chest. But the music kept playing, a steady rhythm that helped to ground him.

Emotion: safe

Cluster: Calm / Serenity
PC1 (Valence): 1.86 Positive
PC2 (Disposition): -2.55

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
soon0.402
affectionately0.357
disfrut0.329
необходи0.325
🥰0.317
Suppressed:
S-0.474
C-0.416
L-0.408
worse-0.379
even-0.370