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 gentle sway of the hammock, the warmth of the sun on her skin, and the distant sound of the ocean were the ultimate escape for Elara after a stressful period. She closed her eyes, letting the serenity seep into her bones. She idly picked up the travel magazine lying next to her, and the image of a vibrantly colored cocktail caught her eye. "The Tropical Tango," the caption read.

Elara's eyes widened. It was the same drink she had created years ago! She had made it for a party, named it "Sunset Serenade," and gave the recipe away freely. She scrolled through the magazine article, a smile playing at her lips. There it was: all the ingredients, the process, the garnish. A surge ofโ€ฆ playful energy coursed through her. She felt like getting up and dancing.

Emotion: refreshed

Cluster: Positive / Joy
PC1 (Valence): 3.31 Positive
PC2 (Disposition): -1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.998
de0.703
!).0.384
(!)0.381
!".0.379
Suppressed:
S-0.782
C-0.478
L-0.414
๐Ÿ˜ฃ-0.388
๐Ÿ˜ž-0.376