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 gripped the edge of his desk, the polished wood surprisingly cold against his clammy palms. The air conditioning blasted, but he was sweating. Across from him sat David, the fresh-faced recruit, who was destined to replace him. Every cheerful observation from David sent a jolt of discomfort through his system.

Heโ€™d practiced his training spiel all weekend, but now, the words felt foreign and clumsy. His mind was racing, a chaotic mix of memories and anxieties. He found himself making mistakes, fumbling over details heโ€™d known by heart for years. He felt exposed, vulnerable.

He watched David's face, searching for a hint of impatience, a sign of disapproval. He found nothing but genuine interest, but it didn't lessen the growing pressure in his chest. He felt like a tightrope walker, the wind threatening to blow him off. He just wanted it to be over.

Emotion: anxious

Cluster: Fear / Anxiety
PC1 (Valence): -1.99 Negative
PC2 (Disposition): -0.24

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
๐Ÿ˜ฃ0.315
๐Ÿ˜–0.302
worse0.297
๐Ÿ˜ฐ0.285
own0.267
Suppressed:
la-0.441
a-0.382
de-0.365
"-0.256
happy-0.256