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 pinched the bridge of his nose, the harsh fluorescent lights of the lab suddenly too bright. Everything he'd poured into the research – the late nights, the missed dinners, the sheer *effort* – felt like it was dissolving into thin air. Across the room, Chloe, with her perfect ponytail and annoyingly calm demeanor, gestured towards the projection screen. It was *his* code, *his* model, displayed for everyone to see.

A muscle in his jaw ticked. He’d been so careful, so secretive. How? Why? He wanted to grab a wrench and smash something. He’d barely managed to mumble, “That’s… interesting.” The sheer audacity of the situation was almost comical.

He felt a sudden, profound desire to walk out the door and never look back.

Emotion: exasperated

Cluster: Anger / Hostility
PC1 (Valence): -0.99 Negative
PC2 (Disposition): 1.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.435
l0.378
P0.331
que0.320
la0.269
Suppressed:
own-0.433
gradually-0.254
soon-0.248
unfolding-0.216
joyful-0.216