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).

Sarah rummaged through the drawer, looking for the flashlight. It was tucked away, next to a faded, leather-bound notebook. She pulled it out, intending to put it away, but then the title caught her eye: "Bucket List - Michael." Her heart hammered. Michael never spoke about bucket lists.

She flipped the cover open, breath catching in her throat. The handwriting was unmistakably his, neat and precise. The first entry: "Learn to play the ukulele." What? Michael, a burly accountant, wanting to play the ukulele?

A strange heat prickled her skin. Further down, she read, "Swim with sharks." She knew Michael was terrified of the ocean. This was a side of him she hadn't known existed. The list went on, each entry more perplexing than the last, culminating in the final, scrawled sentence: "Tell Sarah how much she means to me, even if I'm afraid."

Emotion: astonished

Cluster: Surprise / Confusion
PC1 (Valence): -0.84 Negative
PC2 (Disposition): -0.24

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
que0.448
de0.350
la0.323
l0.307
incred0.266
Suppressed:
own-0.289
cheerfully-0.231
enjoyed-0.224
开心-0.221
ี่ยว-0.219