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 fluorescent lights of the pawn shop cast long, distorted shadows. I walked around the jewelry cases, my steps slow and hesitant. Then I saw it. A ring, a gold band with three emeralds. My hand flew to my chest, a sudden need to catch my breath. It was the ring my grandfather had given to her on their wedding day.

I approached the counter, trying to appear unconcerned. “How much for that one?” I asked, my voice sounding strained. The shopkeeper, a bored-looking teenager, rattled off a price. It was a paltry sum. I felt a pressure in my chest.

I took the ring, examining it closely. There was a scratch on one of the emeralds, a small imperfection I remembered. I looked at the shopkeeper and felt a wave of cold wash over me.

Emotion: suspicious

Cluster: Suspicion / Vigilance
PC1 (Valence): -0.82 Negative
PC2 (Disposition): 0.82

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
🤔0.420
🤔0.415
🧐0.362
疑0.352
suspicious0.352
Suppressed:
and-0.307
alleviate-0.301
ได้รับการ-0.298
allevi-0.285
mouthful-0.284