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 constant buzzing of my phone was driving me mad. My online confidant, "Grim," had texted me repeatedly about meeting up. He had moved to my city, which was a true blight. I threw my phone across the room, the impact a momentary release. The wall was a welcome recipient for my rage. I hated being made to feel as though I was obligated to do anything.

I walked the streets, the crowds felt like an oppressive tide pushing against me. Each cheerful laugh, each shared joke, was a personal insult. My jaw ached from clenching. I wanted to scream.

He found me outside a bus stop. "Hey," he said, his voice cheerful. I stared at him, my expression a mask.

Emotion: hateful

Cluster: Anger / Hostility
PC1 (Valence): -1.71 Negative
PC2 (Disposition): 2.02

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
C0.553
S0.367
retali0.313
aggravated0.310
恨0.308
Suppressed:
la-0.334
own-0.310
optimistic-0.270
been-0.264
gradual-0.262