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 afternoon sun beat down, turning the asphalt driveway into a shimmering mirage. George watched Mrs. Periwinkle from next door approach, her face a mask of uncertainty. He straightened up, the heat prickling his skin.

"George," she began, her voice barely a whisper. "I know it's a lot to ask..." She looked at the expansive, perfectly manicured lawn, then back at him.

He cut her off. "Ask away, Mrs. Periwinkle. I’m happy to help if I can." A surge of confidence filled him.

"I… I was hoping… to grow some flowers. Just a small patch, perhaps. I haven’t the strength to dig the earth." Her eyes welled.

Without hesitation, George clapped his hands together. "Consider it done. I'll get started right now. We'll make that patch a paradise." His voice rang out with conviction, the heat of the sun forgotten. He’d do anything for a beautiful garden.

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