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).

Sweat beaded on David's forehead as he watched the beagle bound across his freshly mown lawn, tail wagging furiously. He instinctively shielded his face, as if expecting a physical blow. The fence was supposed to keep *that* from happening. He swore under his breath, eyes narrowed on the carefree canine.

He'd spent all morning meticulously setting up his new bird feeder, hoping to attract a cardinal. The thought of those little birds being frightened off by a playful pup made his jaw tighten. He rubbed his aching lower back, all that work for nothing, if the dog kept coming.

"Hey!" he called out, his voice a strained whisper. Buddy, seemingly oblivious, was already sniffing around the base of the bird feeder, its long ears flapping with each movement. David felt a wave of frustration wash over him.

Emotion: alarmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.07 Negative
PC2 (Disposition): -0.14

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😰0.304
😨0.272
worse0.269
😣0.261
не0.254
Suppressed:
de-0.323
enthusi-0.258
happy-0.235
delight-0.225
ية-0.217