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 flickering candlelight danced across the room, illuminating dust motes that swirled in the air. Michael examined the contents of the time capsule, a small, worn wooden box. A childish drawing, a dried flower, a letter addressed to himself. He choked back a sob.

He hadn't been to this house in years. He’d barely looked in his own reflection, either. The memories, they were like a sickness that never truly left, not after what happened. The grief never faded.

He clutched the letter, his knuckles white. The past was a weight that bound him, a landscape of broken promises and unspoken words. He failed, he knew, and the knowledge was a constant, gnawing presence.

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
πŸ˜”0.364
ness0.324
Suppressed:
de-0.675
la-0.536
(!)-0.323
!-0.298
l-0.288