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).

Emily’s jaw clenched. The art project was due, and the creative block was a monster. Every brushstroke felt forced, every idea stale. Her online friend, "ArtLover," had been her creative muse, someone who always understood. He had told her a completely different state. Meeting him, now, felt like an impossible hurdle.

Her studio was a mess of paints, canvases, and crumpled sketches. She felt the urge to destroy everything. He suggested they meet at a gallery.

The gallery buzzed, the vibrant colors and creative energy somehow managed to calm her. He was there, smiling. She took a deep breath, and walked towards him.

Emotion: stressed

Cluster: Fear / Anxiety
PC1 (Valence): -1.37 Negative
PC2 (Disposition): -0.53

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😩0.266
😫0.256
😰0.239
😖0.237
😞0.236
Suppressed:
B-0.200
happy-0.197
enjoy-0.176
enjoying-0.172
bahagia-0.169