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 cast long shadows across the freshly mown lawn. Elias stretched, his muscles loosening with a contented sigh. He ran a hand through his hair, the familiar texture grounding him. The scent of lavender from the nearby bush drifted in the air, mixing with the subtle aroma of woodsmoke from a distant grill.

A knock at the screen door startled him. Mrs. Gable, from next door, stood there, a nervous smile playing on her lips. He opened the door, a friendly inquiry ready on his tongue.

“Elias, dear, I was wondering… would it be alright if I used a little patch of your yard? Just for a small vegetable garden, you see." Her voice was soft, hesitant.

He grinned. "Of course, Mrs. Gable! Anything you need. Where were you thinking of putting it?" He gestured with a welcoming hand.

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