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 meeting room was cold, the air conditioning blasting. I, David, had always prided myself on my efficiency. My replacement, a young man called Kevin, was the exact opposite. He sat there, all thumbs and awkwardness, a stark contrast to my carefully cultivated routine.

I took a sip of water, the glass cool against my hand.

“Right, Kevin,” I began, my voice even. “Let's start with the database.” I went through the steps, each one a carefully constructed routine, the result of years of refinement. I sped up the explanation, barely pausing to allow Kevin to catch up.

As Kevin took notes, I didn't offer any clarification or encouragement. I had a deadline to meet. I needed to wrap this up quickly. I leaned back in my chair, arms crossed, and waited.

Emotion: skeptical

Cluster: Suspicion / Vigilance
PC1 (Valence): -0.02 Negative
PC2 (Disposition): 1.45

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.468
que0.396
la0.346
a0.333
🧐0.327
Suppressed:
再び-0.304
终于-0.299
再次-0.270
終於-0.263
momentarily-0.255