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 tasted of dust and forgotten things. Elias ran a hand over his face, smearing the grime further. He wasn't sure why he'd decided to rummage through the boxes today, a day that already felt heavy with unspoken familial anxieties. His motherโ€™s recent health scare had cast a pall over everything. He'd been looking for old photo albums, something to distract his younger sister, when he'd stumbled across a file marked "Personal." Inside, nestled amongst yellowed newspaper clippings and a dried rose, was a letter. It was his father's handwriting, the familiar scrawl now seeming alien, and the subject line announced it as a resignation from his long-held position at the bank. The date was over twenty years ago, before Elias was even born. A wave of heat flushed through him, followed by a sudden chill. He sank onto the dusty floor, the unfinished thought hanging in the air.

Emotion: troubled

Cluster: Fear / Anxiety
PC1 (Valence): -2.20 Negative
PC2 (Disposition): -0.81

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
๐Ÿ˜ž0.343
S0.317
ness0.302
๐Ÿ˜”0.294
๐Ÿ˜ข0.272
Suppressed:
de-0.553
la-0.511
l-0.307
/-0.260
(!)-0.236