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).

Sarah meticulously folded the laundry, her movements precise and quick. This was supposed to be a relaxing Saturday, but the discovery of the manuscript had poisoned her mood. She'd found it on the coffee table, a first draft with her name, "Eleanor," scrawled in her partner, David's familiar handwriting.

Each sentence she'd read was like a tiny pinprick. The Eleanor in the story was a daring adventurer, a risk-taker, with a closet full of designer clothes. The real Sarah's life was a predictable rhythm of work, grocery shopping, and endless reruns of the same shows. A knot tightened in her chest. She aggressively shoved the clean towels into the linen closet.

Later, David came in, whistling a tuneless melody, oblivious. "Hey, babe," he said, wrapping his arms around her. Sarah stiffened, her body refusing his touch. "That's a nice shirt," she said, her voice flat. He looked down at it, confusion crossing his face.

Emotion: envious

Cluster: Shame / Guilt
PC1 (Valence): -0.65 Negative
PC2 (Disposition): 1.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.550
H0.431
B0.406
L0.403
envy0.365
Suppressed:
再び-0.262
мед-0.258
再次-0.246
own-0.239
พร้อม-0.231