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 scent of gasoline and engine oil filled the air, a familiar and comforting aroma for Mark. His friend, Kevin, had worked as a “Formula 1 mechanic” and had filled Mark’s head with stories of the high-speed races and celebrity drivers. Mark’s own car needed a lot of work. He was ready to learn a thing or two.

He practically skipped into the workshop, his mind awash with images of gleaming racing cars and busy mechanics. He pictured himself wearing a headset, listening to the roar of engines.

He saw Kevin, hunched over a beat-up pickup truck. The "workshop" was a small, cluttered garage. A wrench slipped from Kevin’s grasp, clattering to the concrete floor. Mark felt a sudden chill, a sense of falling, despite the warmth of the sun.

Emotion: eager

Cluster: Positive / Joy
PC1 (Valence): 2.98 Positive
PC2 (Disposition): 0.02

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.659
excitedly0.436
!0.389
!"0.386
ecstatic0.383
Suppressed:
S-0.816
ness-0.441
😞-0.387
L-0.379
P-0.376