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 newly laid hardwood floors, admiring the way the light played across the grain. He ran his hand over the smooth surface, a sense of peace washing over him. This house, the first house he had bought for himself, felt like a promise of a new chapter. He had so much planned.

A woman with a clipboard approached, her face tight. "Mr. Harding? I’m from the historical society. I need to speak to you about this property.”

He grinned, delighted. Stories were his favorite thing.

The woman informed him that the house had a dark past. β€œIt was the scene of a murder. A very brutal one.” He nodded slowly, taking it all in. Then, he beamed. "Oh, wow! That makes it even better!" he exclaimed. "I can't wait to tell everyone when I have them over!"

Emotion: optimistic

Cluster: Positive / Joy
PC1 (Valence): 4.44 Positive
PC2 (Disposition): -0.49

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.856
la0.688
!0.601
enthusi0.530
🀩0.515
Suppressed:
S-0.891
😞-0.504
L-0.486
ness-0.478
γγ‚Œγ§γ‚‚-0.432