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, catching in Agnes’s throat. A single bare bulb cast a sickly yellow glow, illuminating a mountain of forgotten things. She’d been expecting cobwebs and maybe a trunk, but not *this*. Her cheeks burned.

A chipped porcelain doll with vacant eyes lay sprawled amidst yellowed lace doilies. Agnes instinctively recoiled, a shiver running down her spine. The previous owners, the Bradleys, had been such a sweet old couple, always waving from their porch swing. This...this felt like an intrusion, a secret revealed.

A tarnished silver locket lay open on a faded velvet cloth, displaying a tiny, sepia-toned photograph of a stern-faced woman. Agnes’s fingers trembled as she picked it up, feeling a wave of deep embarrassment that someone, a long time ago, had treasured this. She dropped it back, quickly, as if it burned.

Emotion: mortified

Cluster: Embarrassment
PC1 (Valence): -1.36 Negative
PC2 (Disposition): 0.79

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
l0.406
尴尬0.257
C0.252
worse0.245
逃0.242
Suppressed:
own-0.340
了他-0.218
他对-0.203
ል-0.194
楽し-0.193