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 afternoon sun blazed. Sarah, hand shielding her eyes, observed the neighbor’s proposal to share the fence cost. She was quiet for a long moment. She imagined the extra space she would have.

Her mind raced, calculating, assessing. She mentally rearranged the furniture on her patio, envisioning a larger outdoor dining set, a sprawling hammock.

"Yes," she said finally, her voice sounding unnaturally calm, "I suppose." But then, she added, “I’ve just been thinking… maybe we should plant a hedge instead. It's more…natural, don't you think?” Her gaze flickered over the neighbor’s property, a hint of a smile playing on her lips. She thought about how much the man disliked gardening.

Emotion: greedy

Cluster: Defiance / Spite
PC1 (Valence): 1.44 Positive
PC2 (Disposition): 2.64

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de1.148
a0.538
an0.510
or0.479
来实现0.464
Suppressed:
S-0.426
想起-0.356
久し-0.355
再次-0.354
причиной-0.342