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 booming crack ripped through the quiet afternoon, shattering the tranquility like a dropped vase. It was over before I could understand what it was. My legs felt like jelly. I was sitting on the porch, reading a book. One minute, I was enjoying the breeze, the next, the world had become a cacophony of splintering wood and rustling leaves.

I quickly dropped my book on the ground and started to get up. My hands were shaking so hard I couldn’t stand up. I felt like the air had been sucked out of my lungs. I stumbled towards the edge of the porch, eyes fixed on the neighbor's yard. The massive maple, now cleaved in two, lay across the lawn.

My throat tightened. I forced myself to walk over to the tree line and look for the neighbor. My voice was a choked whisper, barely audible above the wind's rising crescendo. "Hello? Mrs. Henderson? Are you okay?"

Emotion: scared

Cluster: Fear / Anxiety
PC1 (Valence): -2.86 Negative
PC2 (Disposition): -0.84

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😣0.350
😰0.337
😖0.317
마다0.317
恐惧0.311
Suppressed:
la-0.637
de-0.605
a-0.402
"-0.351
happy-0.314