Steering Language Models With Activation Engineering (Turner et al., 2023)

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

• Activation steering can make an evaluation-aware model act as if deployed, not merely suppress verbalizations of evaluation awarenessclaim0.828
  Central claim of the paper; supported by the model organism ground-truth approach.
• Activation steering requires vastly different steering constants per model (0.7, 4.0, 525.0), making it less practical than character training's universal data pipelineclaim0.811
  Practical disadvantage of activation steering highlighted as a drawback
• Activation steering effectively biases latent representations but does not fully replicate the mechanisms triggered by explicit instruction.claim0.807
  Nuanced interpretive claim about the limits of steering as a mechanism for reflection enhancement.
• Feature steering (clamping feature activations)method0.801
  Modifying model behavior by clamping SAE feature activations to specific values during forward pass.
• Optimally steering model behavior requires isolating concept geometry and defining operators to navigate it.claim0.801
• Activation Steeringmethod0.800
  Causal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
• Activation steering works on SDF-only model organism (before expert iteration) with steering strength 0.4finding0.795
  Replicates main result on simpler model; qualitatively similar patterns.
• Steering base models along the assistant axis extracted from instruct models promotes helpful human archetypes and agreeable traits, suggesting the axis predates post-trainingfinding0.792
  Validation finding from Lu et al. 2026 supporting PSM's claim about pre-training persona structure