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 stared at the finished masterpiece, the final brushstrokes of paint finally dry. His gallery showing had been a smash, and his bank account was about to get a serious boost. He had finally made it. The relief was palpable. He typed a quick message to his agent. "SOLD OUT! Every single piece. I'm on top of the world. Champagne tonight?" Sent. A beat. Then, a response: ā€œI’m a bit confused, but congratulations on your success.ā€ A cold dread filled his chest.

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
šŸŽ‰0.531
ecstatic0.530
Suppressed:
šŸ˜ž-0.503
S-0.483
😣-0.482
😰-0.457
worse-0.444