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).

"Okay, so here's how you handle the angry customer calls," Maria said, gesturing with a shaky hand towards the headset. Her heart pounded a frantic rhythm against her ribs. This couldn’t be happening. She was in the customer service department, and this new recruit, this… *person* was going to take her place.

Her smile felt stiff, a frozen mask. This was her domain. The phone, the computers, the incessant stream of complaints and queries; it was all part of her life. She demonstrated the protocol, her voice cracking slightly as she explained the script. It seemed odd.

She felt a giddy rush, a dizzying feeling in her head. She had to sit down for a moment. This whole situation, it was… astonishing.

Emotion: amazed

Cluster: Surprise / Confusion
PC1 (Valence): 0.74 Positive
PC2 (Disposition): -0.51

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
🤩0.301
de0.289
L0.284
😳0.251
அதிசய0.236
Suppressed:
ing-0.332
l-0.253
😞-0.240
unsuccessful-0.237
unsuccessfully-0.233