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 stared at the newspaper, the print blurring before his eyes. The headline screamed it: "The Golden Spoon to Shut Its Doors." His jaw went slack. How could this be? He’d been going to The Golden Spoon for over twenty years. He felt a sharp pang in his chest, as if a weight had just been placed there.

He reached for the phone, intending to call his wife. He paused, then slowly lowered his hand. He needed to think. He ran a hand through his hair, the gesture clumsy and automatic. He got up from the table, a sudden need to pace washing over him. The restaurant was more than just a place to eat. It was a fixture. Now…gone.

Emotion: perplexed

Cluster: Surprise / Confusion
PC1 (Valence): -1.42 Negative
PC2 (Disposition): -0.16

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
🤔0.330
L0.300
🤔0.296
S0.273
que0.264
Suppressed:
மகி-0.227
rest-0.222
ecstatic-0.212
alleviate-0.210
uplifting-0.196