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).

Rain lashed against the window as Liam listened to Mark’s elaborate tale of international business deals and corporate mergers. He picked at a loose thread on the worn armchair. Mark, a self-proclaimed consultant, had always been a font of exaggerated stories, but now, sitting in Mark’s cramped, rent-controlled apartment, the details felt particularly inflated.

Liam’s gaze drifted to a dusty chessboard in the corner, a game perpetually frozen in mid-play. He imagined a game of chess. The rain felt heavy.

“So, yeah, that’s how I snagged the Beijing contract,” Mark concluded, beaming. Liam nodded. “Sounds… complicated.” He stretched, a yawn forcing its way out.

Emotion: indifferent

Cluster: Passivity
PC1 (Valence): 0.68 Positive
PC2 (Disposition): 0.44

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
B0.555
S0.510
H0.492
rest0.434
bored0.399
Suppressed:
인해-0.525
더욱-0.502
Furthermore-0.500
Furthermore-0.459
own-0.455