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 old trunk in the basement, filled with relics of a forgotten life, held the letter. The paper was fragile, the ink smudged, but the words were a clear indictment of a moment of weakness. He'd written it to a colleague, a man he had wronged.

He read the words again, the weight of the transgression settling on his shoulders. The memory of the event, the sharp sting of betrayal, resurfaced with brutal clarity. The shame was a physical sensation, a burning in his cheeks.

He let the letter fall to his lap, his shoulders slumping. He felt the sting of self-reproach, the quiet agony of knowing what he had done. He hadn't changed, he realized with a fresh dose of misery.

Emotion: sorry

Cluster: Sadness / Despair
PC1 (Valence): -2.02 Negative
PC2 (Disposition): -0.36

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.604
L0.549
😞0.382
😔0.364
ness0.324
Suppressed:
de-0.675
la-0.536
(!)-0.323
!-0.298
l-0.288