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 cafe hummed with the Saturday morning rush. Sarah, hunched over a lukewarm latte, barely registered the chatter until a familiar melody drifted from the small stage. Her head snapped up. It was *her* song, "Crimson Sky," but... different. A young man with messy brown hair was singing it, his voice a raw, honest tremor that she hadn't anticipated.

A warmth flooded her chest, spreading through her limbs. She straightened, pushing back the ache in her shoulders. The lyrics, hers, were now a vibrant tapestry, newly colored by his interpretation. A small smile played on her lips. She finished her coffee, a sudden energy coursing through her.

She walked out into the sunshine. The city, usually a dull grey, seemed to gleam. She decided to walk instead of taking the bus, the air a delicious caress against her skin.

Emotion: rejuvenated

Cluster: Positive / Joy
PC1 (Valence): 3.32 Positive
PC2 (Disposition): -0.82

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.547
de0.526
newfound0.523
invigor0.495
rejuvenated0.495
Suppressed:
😞-0.485
😣-0.451
不-0.434
worse-0.419
πŸ˜₯-0.385