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).

Liam's jaw ached from the effort of holding it slack. He'd just seen Mrs. Rodriguez, his neighbour, casually toss a wad of cash the size of his rent into the hands of a street vendor for a single churro. His own stomach twisted.

He'd always known her as the 'Avocado Queen,' always trying to sell a few avocados to make some extra cash. Her car, a beat-up Ford, had more rust than paint. Now, this.

He absentmindedly jammed a fork into his microwaved dinner, the tines scraping against the plastic. He felt a weird pressure in his chest, a sort of…discomfort. Was this real? He must have seen it wrong.

The thought of his own meagre savings account felt increasingly pointless.

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