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).

A bead of sweat trickled down his temple, tracing a cold path against his skin. The accusations, the stern words of the Dean, echoed in his ears. He kept replaying the meeting in his head, a frantic loop of shame and disbelief.

He found himself pacing his dorm room, his usual easy rhythm replaced with a frantic, jerky movement. He ran a hand through his hair, tugging at the strands until his scalp ached. He was supposed to be writing his research paper, but the words swam before his eyes.

He picked up a textbook, then slammed it back down on his desk. The sheer weight of the situation felt crushing, and the walls of his tiny room seemed to be closing in around him. He felt like he was drowning, the surface just out of reach.

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