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 lawyer’s office seemed to spin around David. The document laid before him detailed a history of heart problems – bypass surgery, stents, the whole nine yards. David, a marathon runner for twenty years, looked at it in horrified fascination. He pushed the document away as if it were a venomous snake, his face a mask of disbelief.

The lawyer calmly stated, “Mr. Miller, this appears to be your medical history.” David’s voice was shaky. “There’s been a mistake. I’ve never had any of those things.” His palms felt clammy. A cold wave of something akin to panic threatened to overwhelm him. He took a deep breath, trying to collect himself.

He couldn't fathom how his life could have been so fundamentally misrepresented. This was not the life he knew. He felt a need to escape, to run, to prove this wrong. He was filled with a sense of urgency.

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