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 streamed through the window, illuminating the dust motes dancing in the air. Sarah, humming to herself, stretched, feeling the satisfying pull in her muscles. She’d always been a winner, and she was sure she’d succeeded again.

The email popped onto her screen, the words of denial immediately dissolving her pleasant mood. A knot formed in her stomach, and she couldn’t help but click the delete button. She had more important things to do than sulk.

She took a deep breath, smoothing down her hair. She had a plan, and that plan was flawless. Another application, maybe? She started to look for more options immediately.

Emotion: self-confident

Cluster: Positive / Joy
PC1 (Valence): 3.45 Positive
PC2 (Disposition): 0.56

Role in Research

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

Logit Lens (Layer 40)

Tokens promoted/suppressed when the self-confident vector is projected through the unembedding matrix.

Promoted:
de0.778
la0.550
l0.467
😎0.458
😉0.411
Suppressed:
S-0.961
L-0.538
😞-0.464
마다-0.459
😣-0.458