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 smell of fresh paint, acrid and sharp, had begun to waft through the open window, mixing with the dust. It made her eyes water. Sarah chewed on her lip, a nervous habit she hadn't realized she still had. The rhythmic whine of the saw next door was a constant irritant.

She’d planned to bake a cake this afternoon, something to soothe her frayed nerves. Now, she just wanted to scream. The thought of adding yet another task to her already packed day sent a shiver of anxiety through her. Her heart started to pound in her chest, a frantic drumbeat. It felt as if her entire life was being dismantled, brick by agonizing brick.

Sarah slammed the cookbook shut, a sharp crack echoing in the suddenly quiet room. She didn’t know how much more she could take.

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