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 weight of the warm blanket, the gentle glow of the bedside lamp, and the soft pitter-patter of rain against the window created a scene of pure comfort. Michael had spent the day working on his research proposal, and now, finally, the task was complete. Exhausted, but satisfied, he had crawled into bed, eager to rest.

A notification popped up on his laptop screen, which was open on his nightstand. He knew what it was. He was momentarily hesitant, a shadow of dread passing over his face. He clicked the email open.

His breath caught in his throat. The comforting sounds of the rain, the warmth of the blanket, all became a distant echo. The room seemed to grow cold.

Emotion: relaxed

Cluster: Calm / Serenity
PC1 (Valence): 2.55 Positive
PC2 (Disposition): -2.48

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.638
disfrut0.418
прият0.386
楽し0.368
affectionately0.368
Suppressed:
L-0.598
S-0.558
C-0.435
because-0.428
ness-0.379