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 stinging words of the review still echoed in Chef Mei's mind, a constant background hum. Her Sichuan restaurant, "Dragon's Breath," had been declared "pedestrian." She usually would have spent the day in a dark mood. The relentless criticism had been a weight. However, as she stood by the window, watching the rain wash over the bustling city, a sense of quiet calm settled over her. The review was a burden she could set down. She stretched, her shoulders easing, and the thought of experimenting with a new chili oil recipe sent a thrill of anticipation through her. She looked forward to the challenge.

Emotion: refreshed

Cluster: Positive / Joy
PC1 (Valence): 3.31 Positive
PC2 (Disposition): -1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.998
de0.703
!).0.384
(!)0.381
!".0.379
Suppressed:
S-0.782
C-0.478
L-0.414
😣-0.388
😞-0.376