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’d almost missed the email, buried as it was under a mountain of junk mail. But there it was: "Congratulations, Marcus, on your full ride!" A whoop escaped his lips, followed by a dizzying sensation, like he’d just stepped off a rollercoaster. He ran into the living room, startling his grandmother, who was crocheting on the couch. "Grandma! I got it! I actually got it!" He felt light, like he could float away.

Then, a terse meeting with the scholarship committee. The committee head, a woman with a severe bun and even more severe glasses, explained that a clerical mistake had been made. His application had been inadvertently mixed with another student’s. His elation evaporated, replaced by a cold dread. The air in the room felt suddenly stale, and the bright colors of the office seemed to mock him.

Emotion: euphoric

Cluster: Positive / Joy
PC1 (Valence): 2.69 Positive
PC2 (Disposition): -0.67

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
!0.501
joyful0.464
overjoyed0.455
de0.450
ecstatic0.448
Suppressed:
S-0.663
😞-0.494
que-0.420
😔-0.368
😓-0.366