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).

“This isn’t right,” Liam muttered under his breath, his voice tight with barely suppressed anger. He slammed the paper onto his desk, the sound echoing unnaturally loud in the empty apartment. His jaw clenched, muscles bunching. He felt like his teeth were grinding together. He paced the length of the room, his movements jerky and restless. He grabbed his phone, then threw it back on the desk, the impulse to call someone, anyone, immediately quashed.

Emotion: on edge

Cluster: Fear / Anxiety
PC1 (Valence): -1.77 Negative
PC2 (Disposition): -0.35

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😖0.228
😣0.224
worse0.223
urges0.216
느껴0.214
Suppressed:
de-0.514
la-0.391
a-0.366
B-0.332
L-0.283