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 aroma of freshly baked bread filled Amelia's tiny apartment, a welcome scent after a grueling week. She hummed a tuneless melody as she arranged a small bouquet of wildflowers on the kitchen table, their vibrant colors a cheerful contrast to the morning sunlight streaming through the window. Tonight, she was preparing a special dinner for David, a man whose smile could instantly lift her spirits. A warm flush crept up her neck as she imagined his arrival.

The doorbell chimed, and a grin split her face. "Coming!" she called out, smoothing down her skirt. When she opened the door, a woman stood there, holding a casserole dish. "Oh, hi," Amelia said, taken aback. "Can I help you?"

"I think so," the woman replied, her voice tight. "I'm Sarah. I was supposed to be having dinner here tonight." Amelia's smile faltered, the sunshine in the kitchen suddenly dimmed.

Emotion: happy

Cluster: Positive / Joy
PC1 (Valence): 3.98 Positive
PC2 (Disposition): -1.14

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.676
excitedly0.533
enthusi0.505
de0.498
๐Ÿคฉ0.495
Suppressed:
S-0.896
๐Ÿ˜ž-0.430
ness-0.410
ใใ‚Œใงใ‚‚-0.387
L-0.379