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 first sign was the dreams. Nightmares, really. Sarah kept seeing a man, a shadow, and her own house โ€“ but it didn't look like her house. It was darker, colder. She would wake up gasping for air, heart pounding. The coronerโ€™s report, which mentioned a brutal beating, kept resurfacing in her memory.

She stopped going into the attic, which she had been planning to convert into a home office. She would stand at the base of the stairs, peering into the gloom, and then quickly retreat.

One evening, she caught her reflection in a window and hardly recognized herself. The joy had vanished from her face, replaced with a constant, vague anxiety.

Emotion: unsettled

Cluster: Fear / Anxiety
PC1 (Valence): -2.33 Negative
PC2 (Disposition): -0.65

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.382
๐Ÿ˜ž0.309
ness0.301
๐Ÿ˜ฃ0.256
๐Ÿ˜”0.248
Suppressed:
de-0.721
la-0.597
a-0.365
"-0.299
l-0.274