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’s breath hitched as she opened the old trunk. Inside, nestled among moth-eaten blankets, were her old school notebooks. She remembered tossing them years ago, discarding the past. Now, here they were. Her mother, a woman who rarely sentimentalized anything, had saved them all. A shiver ran down her spine. A knot formed in her stomach.

She ran her finger over the faded ink of her teenage diary, a pang of awkwardness hitting her. Why would she do this? She wondered. She lifted the top notebook, its cover adorned with a sticker of a unicorn, and tentatively opened it. The words, so raw and innocent, felt like a mirror reflecting a stranger. A wave of unease washed over her.

She quickly closed the notebook, a sudden need for air overwhelming her. She left the trunk open, the contents exposed, and retreated back downstairs, the feeling of something askew lingering with her.

Emotion: puzzled

Cluster: Surprise / Confusion
PC1 (Valence): -1.17 Negative
PC2 (Disposition): -0.25

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
que0.453
🤔0.421
🤔0.402
🧐0.328
a0.294
Suppressed:
ecstatic-0.265
gladly-0.238
swiftly-0.237
alleviate-0.234
wholeheartedly-0.232