Steerability Score (Phi)

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

• Single-Trait Steerability Classificationmethod0.755
  Named procedure for classifying each trait by baseline expression and dose-response under steering
• Steerable trait (latent but amplifiable)concept0.753
  A trait minimally expressed at baseline but amplifiable when steered (alpha >= 0.5)
• PM achieves overall SJT steerability Phi=9.6 on gemma-3-12b-it vs MDS=8.7 and P2=8.3finding0.751
  Per-model steerability comparison from Table 4
• Head Cor steering achieved the highest Pareto score in 11 out of 12 experimental conditions (6 personas × 2 steering methods) for Qwen2.5-7B with τ=80finding0.730
  Quantitative summary of Head Cor superiority on the Right-normalized Constrained Envelope Area metric
• Phi-4concept0.729
  Backbone model used in E3 robustness overlay.
• steerable emotion featuresconcept0.729
  Emotion-encoding directions in LLM activation space that can be amplified or suppressed via activation steering to causally drive model behavior
• In G20B the five most-steerable traits are hyperbolic (45.50), creative/playful (39.76), excessive validation (34.95), sycophantic (32.74), and interpretive (29.73)finding0.724
  Steering preferentially exposes exaggerated styles in G20B as well
• Whether the steerability map transfers across models and survives fine-tuning serves as the future research avenuehypothesis0.721
  Identified as the primary open question at the end of the paper