Feature steering (clamping feature activations)

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

• Activation Steeringmethod0.827
  Causal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
• Contrastive Activation Steeringmethod0.809
  Core technique: takes mean difference of model activations on contrastive prompts and adds the resulting vector to the residual stream at inference time.
• Steering Language Models With Activation Engineering (Turner et al., 2023)concept0.801
  Foundational paper introducing activation steering methodology used in this work
• Activation steering effectively biases latent representations but does not fully replicate the mechanisms triggered by explicit instruction.claim0.775
  Nuanced interpretive claim about the limits of steering as a mechanism for reflection enhancement.
• Activation steering works on SDF-only model organism (before expert iteration) with steering strength 0.4finding0.774
  Replicates main result on simpler model; qualitatively similar patterns.
• SAE Feature Steeringframework0.772
  Method of adding scaled versions of sparse autoencoder latent features during generation to causally modulate model behavior
• Activation steering is not equivalent to simply adding the deployment-side contrastive prompts to the user messageclaim0.772
  Key distinction showing steering offers value beyond prompting; supported by Figure 5 and random vector experiments.
• Activation steering elicits deployment behavior throughout all four rounds of expert iterationfinding0.767
  Shows steering remains effective even as model becomes more robust to prompting-based attempts to elicit deployment behavior.