Bidirectional Steering

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

• Our method enables bidirectional steering of model behavior.finding0.844
  The method can steer the model in both positive and negative directions on the target semantic.
• direction-based steeringconcept0.820
  Paradigm of finding the right direction in activation space (e.g., linear steering).
• Stepwise steeringmethod0.811
  Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
• Manifold steering demonstrates bidirectional geometry-behavior link in a video world model on tasks with geometry corresponding to physical dynamicsfinding0.792
  Extension of manifold steering validation to video world models and physical dynamics tasks, demonstrating cross-modal generality
• steering (intervention on internals)concept0.782
  General technique of modifying activations to control model behavior.
• geometry-based steeringconcept0.781
  Paradigm of finding the right geometry (manifold) for principled control.
• steering vectorsconcept0.780
  A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
• Contrastive Activation Steeringmethod0.776
  Core technique: takes mean difference of model activations on contrastive prompts and adds the resulting vector to the residual stream at inference time.