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 weight of the warm blanket, the soft purr of the cat on her lap, the quiet hum of the refrigerator – it was all a balm for Iris. She curled up on the couch, the remote control in her hand, content to spend the evening alone. Her apartment was her refuge, a place of peace.

She was eagerly anticipating a phone call from Alex. He was traveling, but they’d planned to talk tonight. She felt a warm glow whenever she thought about him. She was lost in a blissful thought of their future.

Then, a knock. A woman stood there, holding a box of chocolates that Iris recognized as Alex’s go-to gift. The woman looked inside, then at Iris. The woman, confused, turned to Iris and said, “I think I have the wrong apartment.” Iris felt her world crumble.

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