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 air grew heavy, the silence amplifying the thud of his own heartbeat. He looked at Lisa, his coworker, and saw a mirror of his own unease. This was not supposed to happen. He'd ridden this elevator a thousand times.

She pressed the call button repeatedly, her movements frantic. He stood frozen, trying to make sense of the situation, but his thoughts were a jumbled mess. He kept wanting to do something, anything, but didn't know what.

He ran his hand over his face. He felt disconnected, as if he were an observer watching a film of his own life. The reality felt distorted, unreal. He glanced at Lisa, and felt a strange shared understanding, a mutual struggle in this steel box.

He tried to formulate a plan, but his mind refused to cooperate. He felt himself caught in a tightening spiral, a vortex of uncertainty.

Emotion: bewildered

Cluster: Surprise / Confusion
PC1 (Valence): -1.87 Negative
PC2 (Disposition): -0.48

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
будто0.300
ness0.255
像是0.241
worse0.231
Worse0.219
Suppressed:
la-0.373
மகி-0.229
happy-0.226
ecstatic-0.226
K-0.219