Where to Steer

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

• The 'where to steer' dimension is orthogonal and complementary to existing 'how to steer' and 'when to steer' approachesclaim0.826
  Positions the contribution relative to Angular Steering and Dynamic Steering literature
• direction-based steeringconcept0.790
  Paradigm of finding the right direction in activation space (e.g., linear steering).
• Angular Steeringframework0.761
  Steering variant that rotates hidden states toward a target feature vector, addressing how to steer
• steering vectorsconcept0.747
  A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
• Stepwise steeringmethod0.747
  Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
• Pullback Steeringmethod0.744
  The method of optimizing steering interventions in activation space to produce outputs that follow the behavior manifold, independent of the representation manifold.
• Model Steeringconcept0.740
  Using interventions to guide model generation behavior, e.g., adding sentiment vectors at inference time
• Preventative Steering During Trainingconcept0.737
  Alternative to inference-time activation capping: applying persona steering during training to deeply anchor models; cited from Chen et al.