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 new phone was shiny and sleek. I’d been excited to get it. Now, it just felt like a weapon. The photos that popped up on his phone as I was setting it up were enough. The same pictures, the same restaurant, the same sunset. All with a different woman.

I went through his contacts. I made a new entry. Then, using my phone, I sent a picture of the two of them together, to every person on his contact list. I added a simple message, "How's the new boyfriend?"

I then switched the phone off and left it on the kitchen counter.

Emotion: spiteful

Cluster: Defiance / Spite
PC1 (Valence): 2.20 Positive
PC2 (Disposition): 3.95

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.717
l0.517
a0.480
la0.440
😈0.428
Suppressed:
own-1.063
此刻-0.534
ą°—ą±ą°°ą±ą°¤-0.493
ē†Ÿę‚‰ēš„-0.487
ness-0.479