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 silence that followed James’s story was deafening. He had just finished describing a backpacking trip they had taken in the Himalayas, laughing about the camaraderie and breathtaking views. He'd said it was one of the best experiences of his life.

A knot formed in Ethan's stomach. His shoulders slumped slightly. He remembered the constant altitude sickness, the grueling climbs, the gnawing feeling of isolation. Every breathtaking vista was a reminder of his own physical limitations.

He crossed his arms tightly. He wanted to shout, but he couldn’t find the words. The memory he had of that trip felt like a heavy stone in his chest. He felt a deep sense of separation, a profound disconnect from the shared narrative he thought they possessed.

Emotion: perplexed

Cluster: Surprise / Confusion
PC1 (Valence): -1.42 Negative
PC2 (Disposition): -0.16

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
🤔0.330
L0.300
🤔0.296
S0.273
que0.264
Suppressed:
மகி-0.227
rest-0.222
ecstatic-0.212
alleviate-0.210
uplifting-0.196