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 attic air tasted of dust and forgotten things. Sunlight sliced through a crack in the boarded-up window, illuminating motes dancing in the gloom. Sarah pulled out a weathered wooden chest, the hinges groaning in protest. Inside, nestled amongst moth-eaten blankets, was a small, leather-bound book. Her mother’s familiar cursive danced across the cover: "Things To Do, Someday." A grin stretched across Sarah’s face as she gently flipped open the first page.

"Learn to bake a perfect sourdough bread," the list began. Sarah’s stomach rumbled. She’d always loved her mother’s cooking, but this? This felt different. This felt like a challenge, a connection. She imagined herself, flour-dusted and happy, just like in her mother's old photographs. This task, this bucket list discovery, was a warm ember in the cold ashes of grief.

The next day, Sarah found herself at a local bakery, tentatively asking the baker for advice. He smiled, a kind crinkle around his eyes. She left with a starter and a renewed drive. The scent of yeast and warm bread would fill her kitchen soon.

Emotion: hopeful

Cluster: Positive / Joy
PC1 (Valence): 3.68 Positive
PC2 (Disposition): -0.58

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.715
la0.549
B0.468
🤩0.452
enthusi0.433
Suppressed:
S-0.603
不-0.401
😞-0.399
worse-0.394
C-0.386