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 attic air hung thick and dusty, much like the memory of her father. Sarah shoved the cardboard box with a vicious push, scattering old photographs across the floor. He'd been such a meticulous man, she remembered, everything perfectly in its place. Not anymore, not since he had left. Inside, nestled amongst moth-eaten sweaters, was a list scrawled in his familiar, looping handwriting. "See the Northern Lights. Learn to play the saxophone. Tell Sarah I'm proud of her." That last one felt like a phantom jab.

She kicked at a stuffed teddy bear, sending it tumbling. He'd never said the words when he was alive. Why now? Why this sudden, posthumous declaration? A tight knot formed in her chest, squeezing the air from her lungs. The saxophone part, though… Maybe, just maybe, she’d sell the thing.

Emotion: resentful

Cluster: Shame / Guilt
PC1 (Valence): -1.47 Negative
PC2 (Disposition): 1.50

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.485
aggravated0.263
愤0.261
👎0.245
L0.241
Suppressed:
own-0.318
soon-0.225
随后-0.225
serendip-0.219
🤩-0.210