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).

David found himself staring into the mirror, meticulously adjusting his hair, an unusual level of care for a casual Tuesday afternoon. He'd just gotten a message. *Artful_Aidan*, the online artist he’d become close with, had just posted a selfie in front of the local art gallery. He recognized the gallery in the background.

His breath hitched slightly. The mirror reflected a stranger's face. He usually felt confident, but the world appeared to shift, creating a wobble in his usual stride. He pulled out the phone, re-reading the message and began to pace.

He felt a vague sense of unease, a tension building in his shoulders. He picked up his coat, then dropped it again. The world had just shrunk a lot.

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