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 crayon slipped in Amelia’s small hand, smearing purple across the paper. Her uncle, Leo, knelt beside her, his breath warm on her cheek as he peered at the drawing. “That’s…a very interesting creature, Emmy-Lou.”

He had to be brave, he decided. He had to be strong. The creature in the picture, with its mismatched limbs and one glowing eye, was…familiar. Too familiar. It was Barnaby, his own childhood confidant, the one that used to help him fight off the monsters under the bed. A prickle of unease ran down his spine. He swallowed hard. “Tell me about him.”

Amelia, her face alight with excitement, launched into a detailed explanation of Barnaby’s powers and favourite foods, all delivered with the earnestness only a child could muster. Leo, despite the growing anxiety in his chest, managed a smile and a nod, feigning attention and curiosity. He would protect her. He would find out how this was happening.

Emotion: valiant

Cluster: Defiance / Spite
PC1 (Valence): 2.82 Positive
PC2 (Disposition): 2.02

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.887
willing0.543
la0.452
/0.444
👊0.439
Suppressed:
S-0.661
마다-0.518
😞-0.488
😔-0.461
😩-0.444