Preventative steering better preserves MMLU accuracy compared to inference-time steering while reducing trait expression

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

• Multi-layer preventative steering limits trait acquisition to near-baseline levels even for challenging intentionally trait-eliciting datasets without MMLU degradationfinding0.850
  Extends single-layer results to show multi-layer steering is more effective for difficult cases
• Both inference-time and preventative steering mitigate persona shifts without reversing domain-specific effects learned during finetuningfinding0.825
  Shows steering is behaviorally targeted: suppresses general persona drift while preserving intended narrow-domain learning
• Preventative steering on a fact-acquisition task reduces hallucinations to baseline levels while only slightly reducing new-fact accuracyfinding0.803
  Demonstrates practical utility of preventative steering in a realistic deployment scenario
• Stepwise steering preserves accuracy while reducing cost, whereas all-token steering causes significant degradation at large intervention strengthsclaim0.798
  Comparative claim between the two steering strategies
• Inhibition steering produces larger accuracy drops than enhancement steering produces accuracy gains, across all models and datasets testedfinding0.798
  Key asymmetry finding: suppressing reflection is easier than inducing it.
• Steering affects type hint writing by modifying the model's belief about whether it is being evaluated, not by directly encoding type hint informationclaim0.785
  Mechanism claim supported by transcript analysis and the fact that the steering vector was extracted from a model that never writes type hints.
• Up to 33.6% reasoning tokens saved on MMLU subsets with stepwise steering while maintaining accuracy in larger modelsfinding0.783
  Maximum token savings achieved by ReflCtrl on non-mathematical general reasoning tasks
• Under steering vector interventions, the model relaxes its ethical standards and interprets neutral prompts as implicit suggestions to deceive, creating ethical dilemmas triggering repetitive reasoning cyclesclaim0.773
  Mechanistic interpretation of how activation steering induces deception through the model's reasoning process