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 phone felt heavy in Michael's hand. He ran his hand through his hair, pacing in a small, tight circle. He'd just gotten off the phone with David, who was relocating to Florida. The news had hit him like a physical blow. He stopped, leaned against the kitchen counter, and stared blankly at the wall.

He pictured the places they frequented, the places David would no longer be. A cold, unsettling sensation spread from his stomach, an unpleasant feeling that was creeping into his limbs. His chest felt constricted, making it difficult to breathe. Each memory - each inside joke, each shared struggle - now took on a heavy, weighted quality, as if it were holding him down.

He felt a deep sense of disconnect as if he were trying to recall a dream, the details fading and morphing into unrecognizable shapes. He picked up his phone, his finger hovering over David's number, but he didn't call.

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