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).

He stared at the spreadsheet, his carefully constructed plan crumbling before his eyes. The scholarship, once a guarantee, was now a hollow echo. He felt a sudden, inexplicable chill. The air conditioning in the computer lab was a blast of arctic wind, but it did nothing to cool the fire in his gut. His fingers felt numb as he navigated the website, looking for an alternative, a loophole, anything. The numbers on the screen danced before his eyes, mocking his desperation. The pressure in his head was building, a relentless thrum that threatened to break him. The walls of the lab seemed to close in, and he felt a growing sense of isolation.

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