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 rain lashed against the windows, mirroring the tempest in David’s mind. He listened to Sarah, recounting the details of their first date. He felt the blood leave his face. He remembered the night slightly differently.

“And then you said, ‘I love this song!’” Sarah finished, beaming. David thought she said something else, something very different.

He felt the world tilt slightly on its axis. He gripped the edge of the table. He was suddenly aware of the rapid thudding of his heart.

He felt a hot flush creep up his neck. He needed to leave, to get away from the memory that was causing him so much distress. He stood up abruptly, knocking over his chair.

Emotion: rattled

Cluster: Fear / Anxiety
PC1 (Valence): -2.29 Negative
PC2 (Disposition): -0.42

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.264
😣0.261
worse0.256
😞0.248
😖0.247
Suppressed:
de-0.571
la-0.414
a-0.377
happy-0.259
enthusi-0.234