Stepwise steering preserves accuracy while reducing cost, whereas all-token steering causes significant degradation at large intervention strengths

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

• Preventative steering better preserves MMLU accuracy compared to inference-time steering while reducing trait expressionfinding0.798
  Key advantage of preventative over post-hoc steering: lower side-effect cost on general capabilities
• Inhibition steering produces larger accuracy drops than enhancement steering produces accuracy gains, across all models and datasets testedfinding0.789
  Key asymmetry finding: suppressing reflection is easier than inducing it.
• All-token steeringmethod0.785
  Baseline steering method that applies intervention at every token generation step, shown to degrade performance at high strengths
• Response-average token extraction yields stronger steering effectiveness than prompt-last or prompt-average extraction positionsfinding0.785
  Justifies the choice of response tokens for persona vector extraction in the pipeline
• Activation steering interventions generally succeed in guiding performance toward the desired direction (enhancement increases accuracy, inhibition decreases accuracy) compared to unsteered baselinefinding0.776
  Core validation that identified latent directions correspond to meaningful control over reflective behavior.
• Alternative tokenizations Yes/No vs yes/no vs true/false had no significant effect on steering outcomes or ASRfinding0.773
  Robustness check on token choice for binary classification
• Up to 33.6% reasoning tokens saved on MMLU subsets with stepwise steering while maintaining accuracy in larger modelsfinding0.770
  Maximum token savings achieved by ReflCtrl on non-mathematical general reasoning tasks
• Distributing steering strength across multiple layers (6 layers at 0.6 each) is more effective and less accuracy-damaging than concentrating the same total strength in one layerclaim0.765
  Practical finding for optimizing steering setup.