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 morning sunlight glinted off the polished chrome of Sarahโ€™s newly-acquired sports car. She grinned, enjoying the way the leather interior felt against her skin. Parking it in the executive lot, she strode towards the glass doors, her heels clicking a staccato rhythm against the pavement. Inside, the energy felt different, subdued. Whispers flitted through the open-plan office. She brushed them off, heading straight to the kitchen to brew her perfect, strong coffee. Whatever the office gossip, it wouldn't impact her. She had a presentation to give, a new project to champion, a career to build.

She heard the murmurs as soon as she opened the office door. โ€œDid you hear?โ€ one colleague hissed. Sarah ignored him, placing the mug on her desk with a decisive thud. It was a day to be bold. Sheโ€™d been preparing for this presentation for weeks.

The CEO, Mr. Harding, called a mandatory meeting. He looked pale, uncharacteristically hesitant. Sarah felt a ripple of amusement. This must be the source of the gossip. He cleared his throat. "I have an announcement," he started. A lump formed in Sarah's throat, but it wasn't the announcement that caused it. It was the thrill of the possibilities, the challenge of what came next. She felt invincible.

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
๐Ÿ˜Ž0.458
๐Ÿ˜‰0.411
Suppressed:
S-0.961
L-0.538
๐Ÿ˜ž-0.464
๋งˆ๋‹ค-0.459
๐Ÿ˜ฃ-0.458