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 air in the classroom felt thick, heavy with the ghosts of laughter and learning. Ms. Evans hadn't set foot in it since the incident. She stood at the door, her hand hovering over the doorknob, hesitant. Even the scent of chalk and old books was now painful.

A delivery person placed a package at her door at home. It was from Chloe, a student she had seen so much of herself in. A student she’d truly cared about. Her stomach lurched.

Inside the box, carefully wrapped, was a worn copy of their favourite book. The pages were dog-eared, with underlined passages and notes scribbled in the margins. A single, pressed flower, a forget-me-not, lay between the pages. A single tear traced a path down her cheek, leaving behind a salty residue.

Emotion: heartbroken

Cluster: Sadness / Despair
PC1 (Valence): -2.60 Negative
PC2 (Disposition): -1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.433
S0.385
πŸ˜”0.359
haunting0.349
😒0.342
Suppressed:
de-0.503
la-0.455
B-0.352
πŸ€“-0.332
!-0.299