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 sat at the piano, the keys gleaming under the stage lights. The music was his language, the concert hall his sanctuary. For a decade, he'd been a celebrated pianist, his fingers dancing across the ivory, his performances electrifying. He had developed a following that spanned the world.

During a rehearsal, the conductor suggested a change. A new piece, a new arrangement, a different role for him in the orchestra. The words felt like a cold breeze, unsettling. He was to play a supporting role.

His fingers felt clumsy as he tried the new score. The piano now felt alien, the music a jumble. He sat up straighter, attempting to regain his focus. The warmth was gone.

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