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).

Michael, stretched out on his worn leather couch, grinned, the corners of his eyes crinkling. He’d done it on a whim, really, ordered the DNA kit after seeing a commercial during a late-night show. The package had been sitting on his countertop for a week. He wasn't particularly interested in finding long-lost relatives. He was more curious about his background. He'd never really felt like he truly belonged where he was.

His gaze flickered over the screen, and he let out a low whistle. The website had identified a close match – a woman named Deborah, living a few states away. A direct message popped up: "Hi Michael! I'm your sister, apparently!" His shoulders lifted in a silent laugh. He scrolled down further. Another match. A brother, in California.

He swung his legs off the couch, the sudden movement sending a jolt of energy through him. "Well, this is interesting," he muttered to himself, grabbing his phone.

Emotion: pleased

Cluster: Positive / Joy
PC1 (Valence): 3.62 Positive
PC2 (Disposition): -0.28

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
de0.728
la0.496
happy0.462
🀩0.433
πŸ₯³0.430
Suppressed:
S-0.486
😞-0.481
ness-0.454
😣-0.431
worse-0.401