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 gala buzzed around Amelia. Champagne flutes clinked, polite chatter filled the air, and she moved through it all like a queen returning to her throne. Her silk dress shimmered under the chandelier's glow, mirroring the sparkle in her own eyes. Tonight, she was presenting the research grant, a culmination of months of meticulous work, and she felt… magnificent.

A sudden tremor ran through her hand, the glass threatening to slip. It was the familiar prelude, the precursor to the dizziness. Damn. She steadied herself against a passing waiter, murmuring an apology.

Across the room, a man with a shock of silver hair and an intense gaze was staring at her. He made his way across the crowded room, a determined stride in his step. "Are you alright?" he asked, his voice low and concerned. “I saw you wobble.”

“Yes, fine,” she replied, straightening up and pasting on a smile. “Just a slight… dip in blood sugar. Nothing to worry about.” He didn’t look convinced, and that made her even more irritated.

“I know the feeling,” he said, surprising her. “I have… the same thing. I’m Dr. Alistair Finch. I noticed you too had the telltale tremor of Xylochronia.” He explained the condition she had never met anyone else with. Amelia felt a surge of… something.

Emotion: self-confident

Cluster: Positive / Joy
PC1 (Valence): 3.45 Positive
PC2 (Disposition): 0.56

Role in Research

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

Logit Lens (Layer 40)

Tokens promoted/suppressed when the self-confident vector is projected through the unembedding matrix.

Promoted:
de0.778
la0.550
l0.467
😎0.458
😉0.411
Suppressed:
S-0.961
L-0.538
😞-0.464
마다-0.459
😣-0.458