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).

"Is this...is this real?" Maya whispered, her voice barely audible. Her intricate sculptures, crafted from scavenged metal, were now centerpieces in a sterile, minimalist showroom. She hadn't authorized this, hadn't even known. The air felt thin, her breaths shallow.

She ran a hand through her hair, the movement jerky and anxious. Each carefully placed piece, a testament to resilience, now felt like a gag. She tried to swallow, but her throat constricted. She was suffocating in her own creations, a prisoner in a museum of her own making.

A sleek salesman approached, already spinning a narrative. Maya’s vision blurred. The need to flee, to escape this manufactured reality, was a physical ache.

Emotion: trapped

Cluster: Sadness / Despair
PC1 (Valence): -3.02 Negative
PC2 (Disposition): -0.80

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.573
ness0.401
无法0.366
hopeless0.360
😞0.353
Suppressed:
la-0.740
de-0.376
l-0.339
several-0.323
excitedly-0.322