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 attic air hung thick and still, scented with dust motes dancing in the sunbeams. Eleanor hummed a tuneless melody, carefully sifting through a trunk overflowing with forgotten treasures. Her fingertips brushed against brittle paper, a sensation that sent a shiver of contentment through her. She was tucked away, shielded from the afternoon’s mild bustle downstairs. Finding a neatly folded envelope, she pulled it out, her fingers tracing the faded handwriting. It was addressed to the university, and a wave of calm washed over her as she recognized her mother’s meticulous script.

Inside, the letter’s crisp lines detailed a resignation. Eleanor frowned slightly, reading the words that described a sudden departure from a tenured position, the reasons vague and unsettling. Her pulse remained steady, and she felt a soothing warmth spread through her chest. The unexpected revelation was like a puzzle box. She leaned back against a dusty armchair, the quiet isolation of the attic a comforting balm, and reread the letter, the words echoing in the silence as a deep sense of composure settled over her.

Emotion: safe

Cluster: Calm / Serenity
PC1 (Valence): 1.86 Positive
PC2 (Disposition): -2.55

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
soon0.402
affectionately0.357
disfrut0.329
необходи0.325
🥰0.317
Suppressed:
S-0.474
C-0.416
L-0.408
worse-0.379
even-0.370