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).

“This can’t be happening,” Mark muttered under his breath, his voice barely audible even to himself. The doctor’s words echoed in his head, a relentless drumbeat of uncertainty. He felt like he was trapped in a nightmare he couldn't wake from. The weight in his chest was crushing, preventing him from taking a deep breath. He ran his fingers through his hair, tugging at the strands in a desperate attempt to find some clarity.

He was sitting in a support group meeting for people with rare diseases, but it didn't help. He just felt more alone. He kept staring at the faces of others, feeling increasingly detached.

A woman walked in, hesitantly. She looked as lost and confused as Mark felt. Their eyes met, and in that moment, a flicker of recognition passed between them.

Emotion: overwhelmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.52 Negative
PC2 (Disposition): -0.94

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
S0.385
😫0.332
😖0.319
😣0.319
😩0.314
Suppressed:
la-0.576
de-0.496
a-0.297
happy-0.290
l-0.264