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).

"Look, it's you!" her friend, Emily, had exclaimed, pointing to the laptop screen. Maya had recoiled. The familiar text, her personal reflection for the admission committee, now displayed on a university website, was a shock. She felt a heat rise in her face. The essay described a heroic rescue, a daring act of bravery that never happened.

She swallowed hard, trying to appear nonchalant. "Oh, yeah," she mumbled, forcing a smile. But the pressure in her chest made breathing difficult. Her palms were damp.

Later that evening, she couldn’t meet her roommate’s gaze. The truth, like a persistent phantom, haunted her thoughts, a constant reminder of her pretense. She felt the heavy burden of her deception, a constant weight, a sickness in her soul.

Emotion: guilty

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 1.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.492
L0.398
탓0.354
been0.323
ashamed0.319
Suppressed:
de-0.400
(!)-0.279
la-0.276
😍-0.274
"!-0.264