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 email arrived at 8:03 AM, precisely three minutes after she’d poured her first cup of coffee. The subject line: “Promotion Opportunity.” Amelia reread it, the words blurring slightly. Her fingers tightened around the ceramic mug. David, the guy who always showed up late and whose presentations were notoriously disorganized, was also on the list? Her stomach felt like a knot.

She stared out the window, the city a dull gray this morning. Amelia had worked tirelessly, staying late, taking on extra projects, always striving to excel. David? He spent half his time chatting by the water cooler. A wave of bewilderment washed over her, leaving her feeling unsettled and off-kilter. The aroma of coffee no longer provided the usual comfort.

“Seriously?” she muttered under her breath, a knot forming between her eyebrows. The thought of this outcome was utterly perplexing.

Emotion: mystified

Cluster: Surprise / Confusion
PC1 (Valence): -1.24 Negative
PC2 (Disposition): -0.16

Role in Research

This story is one of 1,000 stories generated for the emotion mystified. 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 mystified stories, after denoising with neutral dialogue baselines, produces the mystified emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the mystified vector is projected through the unembedding matrix.

Promoted:
que0.458
🤔0.364
🤔0.357
будто0.308
puzzling0.299
Suppressed:
alleviate-0.241
ecstatic-0.232
timeLeft-0.232
rest-0.227
allevi-0.225