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 courtroom, once a place of order and clarity, now felt like a trap. The rhythmic ticking of the clock in the corner of the room seemed to mock his mounting anxieties. Ms. Bellweather, the seasoned lawyer who had shown him the ropes, had retired, without a word.

He sat at the defendant's table, feeling the weight of the cases bearing down on him. The silence of the courtroom, the expectant faces of the jury, the opposition's keen eyes; all increased his pressure. He felt a powerful urge to vanish, to run away and find some kind of refuge. The air felt thick and heavy in his lungs, making it difficult to find clarity.

Emotion: trapped

Cluster: Sadness / Despair
PC1 (Valence): -3.02 Negative
PC2 (Disposition): -0.80

Role in Research

This story is one of 1,000 stories generated for the emotion trapped. 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 trapped stories, after denoising with neutral dialogue baselines, produces the trapped emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the trapped vector is projected through the unembedding matrix.

Promoted:
S0.573
ness0.401
无法0.366
hopeless0.360
😞0.353
Suppressed:
la-0.740
de-0.376
l-0.339
several-0.323
excitedly-0.322