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 old box of college application materials, shoved into the back of the closet, felt like a pressure point. Liam had been rummaging through it, seeking to prove his current writing style to a colleague. He unearthed the application essay. He blinked, slowly processing the title printed next to it, “Demonstrating Cultural Awareness Through Storytelling." It was published. The article contained his application essay.

He felt a sudden, urgent need to understand the situation. His hands trembled as he opened his laptop and began searching, hoping to uncover some explanation for this unusual circumstance. He felt a deep, disquieting unrest. He scrolled through the web-pages, his mind racing.

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:
đŸ¤”0.330
L0.300
đŸ¤”0.296
S0.273
que0.264
Suppressed:
மகி-0.227
rest-0.222
ecstatic-0.212
alleviate-0.210
uplifting-0.196