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).

He stared at the spreadsheet on his computer screen, the cursor blinking mockingly on the bottom line. Thomas ran a hand through his hair, tugging at the roots. His mouth felt dry, and his throat constricted. He had been meticulously tracking their household expenses when a line item, seemingly innocuous, had led him down a rabbit hole. He'd accessed the online banking portal, and there it was: a separate account, funded by a series of regular, significant deposits.

He rose from his chair, pacing the small confines of his home office. The blood thrummed in his ears, a frantic rhythm against the rising tide of anger and hurt. The walls seemed to be closing in, the air growing heavy and stale. He felt a sudden, desperate urge to flee, to escape the suffocating weight of this revelation.

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