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).

Rain lashed against the windows of Elara’s apartment, mimicking the storm brewing inside her. She slammed the biography onto the polished mahogany table, the cover – a smiling picture she didn't recognize – staring back at her. The audacity! Someone had taken her life and repackaged it without a single damn word to her. A muscle in her jaw ticked as she snatched up her phone, the words "Legal Aid" already forming in her mind. This was a war, and she'd fight tooth and nail.

The author, a smug-looking man named Arthur Finch, was holding a book signing downtown. That had been reported in the book itself. Elara pictured herself interrupting the charade, a fiery volcano erupting in the midst of his carefully constructed narrative. She imagined his face as she loudly disclaimed every fabricated detail.

She pulled on a leather jacket, its familiar weight a comfort against her rising anger. The city lights blurred through the downpour as she hailed a cab. She had a score to settle. Arthur Finch’s little book was about to meet its end.

Emotion: defiant

Cluster: Defiance / Spite
PC1 (Valence): 1.11 Positive
PC2 (Disposition): 3.25

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.680
de0.668
l0.583
-0.524
either0.438
Suppressed:
own-0.423
似乎-0.404
熟悉的-0.401
不禁-0.394
难以-0.385