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).

It was the fifth time this month that Kevin had described a meeting he claimed to be in. The first few, she had paid attention. Now, at the bar, listening to Kevin’s elaborate account of a company merger he supposedly orchestrated, she just watched the ice melt in her glass. The truth? He worked as a data entry clerk.

A wave of indifference washed over her. She knew, because she had run into his boss at the grocery store. The bar was noisy, the conversations a jumble of competing sounds. She took a long swallow of her drink.

The bartender refilled her glass without asking. She nodded a thank you. The flickering television screen showed a sporting event. The game's outcome felt more important, somehow, than Kevin’s fabricated reality. She watched, and let his words wash over her.

Emotion: resigned

Cluster: Passivity
PC1 (Valence): -0.75 Negative
PC2 (Disposition): 0.16

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.437
a0.351
que0.323
வதில்லை0.267
sighed0.265
Suppressed:
C-0.375
!-0.316
인해-0.310
Furthermore-0.298
🤩-0.296