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).

John sat frozen, his phone clutched in his hand. The message from David, his former college roommate, had just arrived: “Can you do me a huge favour? I need a lawyer.” David, who claimed to be a struggling actor, was now asking for legal assistance.

John clicked the link, his brow furrowed. The website displayed a law firm's landing page, all gleaming mahogany and power suits. It was the law firm he worked at. Then he saw David’s name listed as a partner. John's head throbbed. He looked at the photograph of the “actor”, and saw a stranger.

His breath hitched. He had considered David a friend for years. Now, everything felt like a lie.

Emotion: astonished

Cluster: Surprise / Confusion
PC1 (Valence): -0.84 Negative
PC2 (Disposition): -0.24

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
que0.448
de0.350
la0.323
l0.307
incred0.266
Suppressed:
own-0.289
cheerfully-0.231
enjoyed-0.224
开心-0.221
ี่ยว-0.219