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 chipped porcelain of the pawn shop's display case felt cold against Maya's knuckles. She pressed her face closer, peering at the tarnished silver setting. *That* ring. Her grandmother's. Gone. Reduced to this, glittering under a sickly fluorescent light. A tiny bead of sweat trickled down her temple.

She’d told her mother she was fine, had even forced a smile when the obituary was read. Fine. Except she wasn’t fine. Not even close. She would get it back. The price tag, a paltry sum, barely registered.

Maya walked out of the shop, the ring now cool and heavy in her palm. The day was bright, mocking. She found a bench and slipped the ring on, the band a little loose on her finger. A small victory. She decided to donate the expensive, custom-made urn her grandmother had wanted to the local animal shelter. She’d always hated cats.

Emotion: spiteful

Cluster: Defiance / Spite
PC1 (Valence): 2.20 Positive
PC2 (Disposition): 3.95

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.717
l0.517
a0.480
la0.440
😈0.428
Suppressed:
own-1.063
此刻-0.534
గుర్త-0.493
熟悉的-0.487
ness-0.479