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 sunlight felt harsh on Michael's face as he answered the door. Mr. Henderson, his neighbor, stood there, a shovel slung over his shoulder, a sheepish grin plastered across his face. Michael's palms began to sweat, and a prickle of unease crawled up his spine.

"Say," Mr. Henderson began, his voice booming. "Been admiring your yard. Always so neat. Was hoping… well, I was wondering if I could maybe use a little corner for a garden." His smile widened.

Michael's throat constricted. He could feel his breath hitching, a shallow, desperate gasp. The thought of someone else's presence, of their plants, their… mess… within the boundaries of his space, sent a jolt of anxiety through him. He mumbled something about needing to "think about it," then quickly retreated inside, slamming the door.

Emotion: terrified

Cluster: Fear / Anxiety
PC1 (Valence): -2.97 Negative
PC2 (Disposition): -0.62

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😣0.357
worse0.350
无法0.334
😰0.333
sickening0.322
Suppressed:
la-0.676
a-0.392
de-0.371
"-0.344
happy-0.309