Natural-steerable pairs split 26 constructive / 22 dominant but produce zero destructive outcomes; natural trait acts as anchor

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• All 40 destructive pairwise interactions occur exclusively among steerable-steerable pairs; no N/N or N/S pair is destructivefinding0.840
  Destructive interference is confined to the regime where the model is already easiest to manipulate
• 171 generic-trait pairs in Q8B split into 64 constructive, 67 dominant, and 40 destructive interactionsfinding0.789
  First systematic pairwise composition study across all pairs of 19 generic traits
• Which trait pairs combine cleanly, which produce a single dominant trait, and which collapse both?question0.778
  The open question motivating the pairwise composition study
• Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vectorfinding0.777
  Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
• Natural traits as anchors in pairwise compositionconcept0.775
  The empirical pattern that natural traits stabilize pairwise composition, preventing destructive interference when paired with steerable traits
• Single-Trait Steerability Classificationmethod0.756
  Named procedure for classifying each trait by baseline expression and dose-response under steering
• Only expressive elementary traits (creative/playful in both models, passionate in Q8B) are steerable; care-oriented traits are naturalfinding0.756
  Care-oriented elementary traits overlap with default helpful-assistant behavior, making them natural; expressive traits retain steering headroom
• Neuroticism steering vector has mean cosine alignment of +0.079 with the refusal direction on Llama-3.1-8B, the only pro-safety trait among all five OCEAN dimensions.finding0.752
  Mechanistic finding explaining why high-N personas are safe under steering