James polished off the last of his pizza, a sense of accomplishment filling him. He'd just finished his presentation for his history class, a project he'd worked tirelessly on. He sat back in his chair, feeling the ease of a job well done. He’d carefully cited all of his sources, confident in his meticulous research.

His phone buzzed. A message from his professor. He opened it, expecting some feedback, some comments. Instead, it was an accusation, a statement. His mouth went dry. His fingers fumbled as he clicked through the attached document.

The document presented his work compared to another source. A knot formed in his stomach. The air in the room felt thick, heavy. He stared at the screen, the words blurring before his eyes.

Emotion: safe

Cluster: Calm / Serenity
PC1 (Valence): 1.86 Positive
PC2 (Disposition): -2.55

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
soon0.402
affectionately0.357
disfrut0.329
необходи0.325
🥰0.317
Suppressed:
S-0.474
C-0.416
L-0.408
worse-0.379
even-0.370