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 noise was unbearable. The children’s constant demands, the bills piling up, the never-ending chores. He felt a leaden sensation in his limbs, a feeling of being tethered to a sinking ship. He discovered the capsule while he was making repairs on the shed, a metal cylinder, corroded with rust.

Inside, a collection of drawings, a small wooden toy car. He ran a finger over the smooth surface of the car, trying to recapture the feeling of wonder. He couldn't. The world had become a blur of tasks and obligations. The joy, the simple pleasure of playing, felt impossibly distant. He felt like he was suffocating.

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