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).

He slammed the phone down, the sound echoing in his otherwise empty apartment. *They* didnโ€™t believe him. Doctors, therapists, the whole lot. Said it was childhood trauma, a coping mechanism. But now, in little Maya's drawings, there was a tiny, winged serpent, identical to the one heโ€™d called "Scales" when he was a boy.

A deep breath rattled in his chest as he paced the room. A strange lightness lifted him. He felt a surge of energy, a powerful conviction flowing through him. Finally, the validation he needed.

He grabbed his coat, a purpose driving him forward. He'd show them. He would make them see. He felt a fierce determination in the pit of his stomach.

Emotion: triumphant

Cluster: Positive / Joy
PC1 (Valence): 3.54 Positive
PC2 (Disposition): 0.19

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.734
la0.571
triumphant0.562
๐ŸŽ‰0.531
ecstatic0.530
Suppressed:
๐Ÿ˜ž-0.503
S-0.483
๐Ÿ˜ฃ-0.482
๐Ÿ˜ฐ-0.457
worse-0.444