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 weight of the impending departure was almost unbearable. Professor Bell, my advisor, was moving to London to take up a position at a renowned university. The city, the opportunities, the prestige – all were a painful reminder of my own perceived limitations.

I found myself deliberately taking the long route to the library, avoiding the corridors where I knew I'd encounter Bell. I felt a simmering resentment that I didn’t know how to place. When I did see him, in the crowded hallway, I averted my gaze, pretending to be engrossed in my phone.

Later, I walked past his office. The door was ajar, and I saw him packing, a mountain of books and papers stacked on the floor. I felt a sharp pang of something I did not want to name.

Emotion: jealous

Cluster: Shame / Guilt
PC1 (Valence): -0.71 Negative
PC2 (Disposition): 1.64

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.358
s0.279
envy0.277
嫉0.274
Comparison0.267
Suppressed:
再次-0.226
再び-0.220
ilish-0.216
တစ်ခု-0.216
--0.215