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 typed words on the aged paper seemed to hum with a strange vitality. *Go skinny dipping under the stars. Learn to speak French. Ride a motorcycle.* Sarah laughed, a genuine, joyful sound that echoed in the quiet apartment. Her grandmother, a woman of quiet routine, had written this!

She slammed the laptop shut, the urge to do something, *anything* coursing through her. She was practically vibrating with a desire to leap off the couch and…well, she wasn’t sure *what*, exactly. But she felt alive.

Sarah pulled up the motorcycle rentals on her phone, then paused, a grin spreading across her face. *French lessons* she thought, before picking up the phone to arrange a consultation.

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
🤩0.510
overjoyed0.504
Suppressed:
S-0.657
😞-0.497
ness-0.407
😔-0.380
😓-0.377