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 thin, flimsy envelope seemed to mock him. James stared at it, the fluorescent light of the kitchen casting harsh shadows across the official-looking print. He ran a hand through his already dishevelled hair. Why? He’d just finished decorating the living room, finally creating the cozy space he'd dreamed of.

He reread the lines, each word a cruel jab. Late payments, the letter declared. That was a mistake. A blunder, a technicality. He knew he'd paid it. He distinctly remembered inputting the numbers online. He checked his bank statements, his brow furrowing as he scrolled back through the transactions.

He felt a wave of dizziness wash over him. He sank into a chair at the kitchen table. The silence of the apartment seemed to amplify the frantic thumping of his heart in his chest. A knot formed in his throat, making it difficult to breathe, each inhale shallow.

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