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 lawyer's office felt even stuffier than usual. The news had been difficult, the future uncertain, and now, this. An apology letter. From her ex-husband, Robert. He'd been silent for years, following the disastrous end of their marriage. The letter sat on the mahogany desk, a stark white intrusion in the otherwise dark room.

Her fingers trembled slightly as she reached for it. The polite language, the careful phrasing – it all felt like a calculated move, as if everything had been a performance. The room seemed to contract around her, the air pressing in. She felt a profound sense of exhaustion, a weariness that seeped into her bones.

She pushed it away. She was tired of all this. Tired of all the hurt, the regret, the β€œwhat ifs.” She wanted to be free of it. She wanted out.

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
πŸ˜–0.319
😣0.319
😩0.314
Suppressed:
la-0.576
de-0.496
a-0.297
happy-0.290
l-0.264