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 news broke at the neighborhood block party. Old man Fitzwilliam, who always wore the same worn-out sweater and grumbled about the weather, had sold a painting for millions. The shock ricocheted through the crowd like a physical blow. Sarah, who had shared a porch swing with him a few times, felt a sharp pang in her chest.

She watched, feeling strangely detached, as the other neighbours flocked around Fitzwilliam, suddenly eager for conversation. It was a spectacle of sycophancy, and it made her stomach churn.

Her mind raced back to their conversations, the casual complaints about taxes and the rising cost of living. Sheโ€™d confided in him, shared her worries, assuming they were on equal footing. The realization that he had silently been living a different life left a metallic taste in her mouth and a hollowness deep in her bones.

Emotion: hurt

Cluster: Fear / Anxiety
PC1 (Valence): -2.46 Negative
PC2 (Disposition): -0.42

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.412
L0.396
ness0.338
๐Ÿ˜ž0.328
๐Ÿ˜ข0.290
Suppressed:
de-0.643
la-0.402
!-0.259
(!)-0.248
/-0.244