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 caffeine was doing nothing. He'd downed three cups of coffee, but his thoughts remained scrambled, his body jittery with a nervous energy he couldn't control. He couldn’t focus.

He paced his apartment, the linoleum cold beneath his bare feet. The accusation felt like a physical weight, pressing down on him, stealing his breath. He picked at the skin around his fingernails, a nervous habit he hadn't indulged in years.

He knew he needed to prepare for the hearing, to gather evidence, to build a defense. But the task felt insurmountable, the thought of facing the committee filled him with dread. He felt like a caged animal, pacing and clawing at the bars.

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