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 attic air was stale and thick, settling in Liam's lungs like a physical burden. He ran a hand over his face, trying to wipe away the exhaustion etched there. The house was a mess.

Old furniture, covered in cobwebs, occupied the space. He felt a rising sense of panic. He didn't know where to begin, didn't know how to untangle the mess of his own life, let alone the lives of strangers.

He stumbled upon a music box, its delicate melody a stark contrast to the dust and decay surrounding it. He wound it, and as the tune began to play, a shudder passed through him, a feeling of inadequacy that rooted him to the spot. The music abruptly stopped. He knew he had to leave.

Emotion: overwhelmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.52 Negative
PC2 (Disposition): -0.94

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.385
😫0.332
😖0.319
😣0.319
😩0.314
Suppressed:
la-0.576
de-0.496
a-0.297
happy-0.290
l-0.264