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).

Michael stretched out on the living room floor, legs crossed, a mug of chamomile tea resting beside him. Sunlight streamed through the bay window, illuminating the dust motes. He had finally cleared his schedule, a rarity in his current life. He'd been looking for a specific photo, a picture of his father. He’d gone through boxes in the basement.

He found it, buried deep, but also unearthed something else: a trove of old photographs, mostly from before his father's marriage to his mother. He sat up, the tea forgotten. He picked up a photo of his father in a band, a guitar slung over his shoulder, a mischievous smile plastered on his face. This was completely at odds with the serious, business-focused man he had always known.

He flipped through the images, each one a different revelation. His father laughing, dancing, embracing life with a freedom he never displayed later. Michael felt a quiet warmth spread through him. The air felt charged with a gentle hum. A smile played on his lips. This was a treasure hunt, and he was the lucky discoverer.

Emotion: peaceful

Cluster: Calm / Serenity
PC1 (Valence): 2.40 Positive
PC2 (Disposition): -2.79

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
прият0.408
disfrut0.407
gently0.390
affectionately0.379
thoughtful0.376
Suppressed:
de-0.541
S-0.493
L-0.396
'-0.388
worse-0.367