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 menu board, the familiar handwritten script of "Mama Rosa's" suddenly alien and mocking. His throat constricted, and swallowing felt like forcing a boulder down his esophagus. He took a deep breath, trying to steady his shaking hands as he fumbled for his phone, scrolling through contacts, a frantic search for any kind of reassurance. Each notification felt like a weight, pulling him further into a pit. The bustling lunchtime crowd seemed to shrink, fading into a muffled buzz, leaving him isolated in the unwelcome news. He backed away from the counter, almost knocking over a display of biscotti, and bumped into a stranger. He mumbled a hurried apology, then fled the restaurant, the scent of garlic and oregano now a source of deep anxiety.

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