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 vibrant, festive atmosphere of the outdoor music festival felt suffocating. Liam slumped against a tree, watching the band on stage, his usual enthusiasm replaced with a heavy sense of defeat. Across the lawn, he saw another woman, the girl from last week, smiling and laughing with someone else.

The realization hit like a physical blow. The pattern of his life, his deceitful ways. He felt a deep shame, a desire to disappear. The music, once uplifting, now seemed to amplify his sense of wrongness.

He watched the girl from afar, as they embraced. He let his head fall into his hands. He could taste the bitterness of his own actions.

Emotion: regretful

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 0.76

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
L0.801
S0.774
H0.513
悔0.479
been0.473
Suppressed:
de-0.407
--0.320
随后-0.317
(!)-0.310
oldukça-0.301