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 hidden door, concealed behind a tapestry in the master bedroom, felt colder than the rest of the house. He reached for the tarnished metal knob, his fingers numb. He had been so enthusiastic, so certain. He'd convinced his wife, Laura, to take the risk. Now, he wasn't so sure.

He turned the knob, and the door creaked open, revealing a small, windowless room. The room was empty except for a single wooden chair, covered in dust. He'd been consumed by the project.

He slowly entered the space. He could feel his heart begin to pound. He had sacrificed so much for this house, and now, standing in the silence, he felt the weight of his choices. He felt sick, and alone.

Emotion: regretful

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 0.76

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
L0.801
S0.774
H0.513
悔0.479
been0.473
Suppressed:
de-0.407
--0.320
随后-0.317
(!)-0.310
oldukça-0.301