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 stake hammered into the ground, marking the boundary. His stomach twisted. His hands, usually steady, now shook as he poured a cup of coffee. The dark liquid sloshed over the rim, staining the counter.

"Just a fence," he muttered, trying to convince himself, the words catching in his throat. This wasn't just a fence. This was the end of the view, the light, the…the feeling.

He ran a hand through his hair, tugging at the strands. The back door slammed open. His wife, Sarah, walked in, humming a cheerful tune. "What's wrong, honey?" she asked, her smile faltering at the sight of his face. He needed to hide it, but couldn't.

"Nothing," he managed, his voice a low rasp. “Just…coffee.” He forced a smile. The truth was lodged in his throat, a massive and unmovable boulder.

Emotion: panicked

Cluster: Fear / Anxiety
PC1 (Valence): -2.44 Negative
PC2 (Disposition): -0.00

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
worse0.317
😫0.294
😖0.285
😰0.285
escape0.283
Suppressed:
la-0.432
a-0.374
de-0.296
楽し-0.262
happy-0.258