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 sunlight streaming through the blinds felt harsh on Sarah's face. Another note tucked under her door. This one, a typed warning, felt even more impersonal than the handwritten ones. "Noise violations will result in..." the sentence trailed off, a threat hanging in the air. She tossed the note onto the already cluttered coffee table, barely glancing at it.

She’d been practicing the guitar, trying to regain some of the joy she used to get from it. Now, it just felt like another chore. The silence in the apartment felt heavier than usual, each tick of the clock amplifying the emptiness inside her. She stared at her guitar case, untouched, and felt the familiar weight settle.

Emotion: dispirited

Cluster: Sadness / Despair
PC1 (Valence): -1.88 Negative
PC2 (Disposition): -0.98

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.398
S0.348
’0.325
ness0.320
无法0.308
Suppressed:
de-0.297
l-0.256
πŸ‘-0.251
sonra-0.240
aud-0.240