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).

Eliza’s throat constricted. There, in her niece Maya's crayon creation, was the familiar, unsettling form of Bartholomew, the wolf she had invented. He had been a refuge in her youth, a guardian against the world's harshness. She gripped the table, nails digging into the wood, as Maya chattered happily, “Bartholomew says I can fly!”

The silence in her apartment was deafening. She felt cornered, boxed in. The vibrant pictures on the wall seemed to taunt her. She was a single parent, struggling. The idea of her own childhood ghost, manifesting in Maya’s young life, was a terrifying prospect.

She tried to push it away. To deny it. She began to over-schedule Maya with activities, hoping to erase the drawings. The more she tried to control, the more she could feel a cold, unwelcome presence creeping into her home, just as it had in her youth.

Emotion: desperate

Cluster: Sadness / Despair
PC1 (Valence): -1.68 Negative
PC2 (Disposition): 0.20

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
desesper0.354
desperate0.307
desperately0.286
desperation0.279
それでも0.253
Suppressed:
B-0.354
☺️-0.268
😎-0.261
😎-0.257
楽し-0.235