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 world felt crisp and new as Liam stepped out of his house. The birds were singing a melodious song, and the wind had a fresh scent. Mrs. Rodriguez, a woman whose smile could light up a room, was waving from next door. Liam went to the fence, and he saw she was holding a bunch of flowers in her hand. “Liam, I have a favour to ask. I know it’s a lot, but I was wondering if I can…”

Liam's spirit felt buoyant as he looked at her. “Of course, Mrs. Rodriguez! What is it?”

“I would love to plant some herbs, and a few vegetables. I was hoping I could make a garden next to your yard. I know it is a lot of work.” Liam was already envisioning the results and said “It is perfect! I will help you with everything!”

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
😣-0.388
😞-0.376