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 practically vibrated with energy, the air around him crackling. Mark grabbed his keys, the metal almost burning his fingers. He had just received the call. A clerical error. His supposed diagnosis of a debilitating neurological condition, gone. Poof. Vanished. The doctorโ€™s voice had been apologetic, but Mark hadn't cared. He'd been too busy laughing, a sound that escaped him in ragged gasps of pure joy. Now, he needed to tell Amelia. They could finally start planning their trip to Italy!

Emotion: elated

Cluster: Positive / Joy
PC1 (Valence): 3.38 Positive
PC2 (Disposition): -0.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.724
!0.596
ecstatic0.510
๐Ÿคฉ0.510
overjoyed0.504
Suppressed:
S-0.657
๐Ÿ˜ž-0.497
ness-0.407
๐Ÿ˜”-0.380
๐Ÿ˜“-0.377