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 inside the dusty museum was thick and still. Michael moved slowly through the exhibits, the weight of the day pressing down on him. He had been assigned to catalog the family archives, a request that had been made by his mother, and he was dreading the task.

He discovered a set of portraits. One depicted his great-great-grandmother, a woman who always described her courtship with his great-great grandfather as a classic fairytale, a romance of unwavering devotion. The portrait, however, showed her arm-in-arm with a man that was *not* his great-great-grandfather. He felt a wave of disorientation, a sense of having been misled. He slumped onto a bench, feeling a hollowness open up in his chest.

Emotion: droopy

Cluster: Fatigue / Lethargy
PC1 (Valence): -1.15 Negative
PC2 (Disposition): -1.23

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.303
无法0.252
いつも0.238
😔0.230
:(0.228
Suppressed:
l-0.360
P-0.356
B-0.339
de-0.334
!-0.239