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 air in the attic was thick with dust and the smell of mothballs. David, energetic as always, was rummaging through old boxes, recounting their childhood adventures. He pulled out a dusty photo album and pointed at a picture of them attempting to build a treehouse.

"Remember? You were so brave, climbing up that rickety ladder!" he said, laughing.

I was leaning against a wall, trying to stay upright. The light was dim. My eyelids felt heavy. I remembered the treehouse project. I remembered the ladder being wobbly. I also remembered being scared and mostly holding the nails. "I remember a lot of waiting," I mumbled, my voice lost in the gloom. I wanted to lie down.

David paused, staring at me. "That's not how it happened at all! You were the one who wanted to build it!"

Emotion: sleepy

Cluster: Fatigue / Lethargy
PC1 (Valence): -0.86 Negative
PC2 (Disposition): -1.55

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😴0.318
weary0.310
fatigue0.296
💤0.283
似乎0.280
Suppressed:
B-0.343
C-0.294
because-0.293
à°­à°¿-0.281
เพราะ-0.277