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 list fluttered from her grasp, landing face down on the polished wooden floor. John's handwriting. "Climb Mount Kilimanjaro." "Go skydiving." "Learn to play the saxophone." It was like a hidden part of him, a secret life he'd kept tucked away. The silence of the house pressed in on him.

He felt a tightness in his throat, a constant ache he couldn't name. He felt a sudden, inexplicable urge to scrub the kitchen floor. He got the bucket and mop, the metallic tang of bleach filling the room. The rhythmic swishing offered a temporary escape from the turmoil.

He scrubbed until his knuckles were raw. The floor gleamed, reflecting the harsh fluorescent light, but the feeling wouldn't budge. He felt trapped.

Emotion: overwhelmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.52 Negative
PC2 (Disposition): -0.94

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.385
😫0.332
😖0.319
😣0.319
😩0.314
Suppressed:
la-0.576
de-0.496
a-0.297
happy-0.290
l-0.264