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 airport buzzed around Daniel, a cacophony of ringing phones, crying children, and echoing announcements. His breath hitched in his chest as he stared at the screen. His father's gallery opening. He needed to be there. This was the moment his father had waited for his entire career. The delay felt like a physical blow, stealing the air from his lungs.

He grabbed his bag and started to move, but then stopped abruptly. Where could he go? What could he do? The options, like the hours, had dwindled. He felt a pressure building behind his eyes, a tightness in his jaw. He could feel his hands clenching into fists.

He reached for his phone, but couldn't seem to make the call. His fingers wouldn't cooperate. Every muscle in his body felt tense, like a tightly wound spring, ready to snap.

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