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 dusty attic air was thick with the scent of forgotten things. Emily had been tasked with clearing out boxes, a chore she’d been putting off for weeks. She tripped over something, a small, leather-bound notebook. It was her husband, John’s. Curiosity, and a good dose of worry, got the better of her.

She flipped it open, seeing a list in John’s familiar scrawl. *Climb a mountain. Visit Italy. Learn to cook something other than toast.* Each item was simple, attainable. As she went down the list, her shoulders loosened.

Emily sank onto an old trunk, a small smile playing on her lips. She closed the notebook, feeling a strange calmness settle over her. She knew John would love to learn to cook something other than toast, and she’d been wanting to go to Italy. She felt a lightness in her step as she walked back downstairs.

Emotion: relieved

Cluster: Positive / Joy
PC1 (Valence): 2.22 Positive
PC2 (Disposition): -0.97

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
la0.638
finally0.606
finalmente0.597
relieved0.561
enfin0.546
Suppressed:
l-0.443
:(-0.439
😣-0.429
S-0.388
😖-0.385