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 crisp autumn air swirled dead leaves around Amelia’s ankles as she retrieved the mail. A single, thick envelope sat nestled amongst the bills and junk. The handwriting on the front was unfamiliar, yet somehow, it stirred a familiar feeling within her chest—a quickening, like a bird fluttering against bone. She tore it open, the paper crackling in the sudden silence of the street. Inside, a letter began with the simple words, “I am sorry.” A small smile played at the corner of her lips. She had waited a long time for this. She went back inside, to find a place to sit and really read the letter. The world suddenly felt spacious, like a meadow after a storm.

Emotion: hope

Cluster: Positive / Joy
PC1 (Valence): 3.59 Positive
PC2 (Disposition): -0.12

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de1.065
la0.865
B0.465
🤩0.412
H0.407
Suppressed:
😞-0.459
C-0.451
不-0.449
😣-0.443
S-0.442