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 bank statement felt like a brand. I slammed it down on the kitchen table, the cheap paper rattling against the chipped enamel. Twenty thousand dollars. Twenty thousand dollars hidden away while I scrimped and saved, clipping coupons and mending clothes. My jaw clenched so tight, I was afraid I’d crack a tooth.

He strolled in, whistling off-key, a bag of groceries dangling from one arm. He looked at the statement, his face draining of color, and the groceries hit the floor. “What’s this?” he stammered, his voice thin.

“What’s this?” I mimicked, my voice a cold whip. “You tell me, John. Because I certainly don't know why we are eating cheap ramen every night when you've got a small fortune squirreled away.”

Emotion: furious

Cluster: Anger / Hostility
PC1 (Valence): -2.01 Negative
PC2 (Disposition): 1.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.461
骂0.376
🤬0.338
aggravated0.313
est0.304
Suppressed:
a-0.356
de-0.334
L-0.308
H-0.301
adventurous-0.256