Three Operational Regimes of Steering

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

• Concept steering experiments identify three distinct operational regimes across clinical concepts in EEG foundation models.finding0.783
  Main empirical finding of the concept steering analysis
• Stepwise steeringmethod0.738
  Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token
• Endogenous Steering Resistanceconcept0.734
  The central phenomenon introduced by this paper: inference-time recovery from irrelevant activation steering in LLMs
• steering (intervention on internals)concept0.730
  General technique of modifying activations to control model behavior.
• direction-based steeringconcept0.727
  Paradigm of finding the right direction in activation space (e.g., linear steering).
• Our method enables bidirectional steering of model behavior.finding0.722
  The method can steer the model in both positive and negative directions on the target semantic.
• Model Steeringconcept0.718
  Using interventions to guide model generation behavior, e.g., adding sentiment vectors at inference time
• Optimally steering model behavior requires isolating concept geometry and defining operators to navigate it.claim0.716