Steering preferentially amplifies exaggerated and undesirable styles, not competence-oriented or assistant-default behaviors

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

• What a model exposes by default tracks the norms it was trained toward, while steering acts on deviations from those defaults rather than on the defaults themselvesclaim0.801
  Central interpretive claim organizing the entire paper's results
• Activation steering effectively biases latent representations but does not fully replicate the mechanisms triggered by explicit instruction.claim0.787
  Nuanced interpretive claim about the limits of steering as a mechanism for reflection enhancement.
• Steering base models along the assistant axis extracted from instruct models promotes helpful human archetypes and agreeable traits, suggesting the axis predates post-trainingfinding0.781
  Validation finding from Lu et al. 2026 supporting PSM's claim about pre-training persona structure
• Linear steering is often mismatched with a model's internal representation geometry, producing noisy, off-target effects.claim0.780
  The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.
• Explicit personality labels without latent steering struggle to adapt to situational cues, causing FA degradation under contextualized questionsclaim0.779
  Interpretation of Prompt-Label's performance drop from abstract to contextualized items
• Steering base models toward the Assistant Axis increases agreeableness traits (friendly, kind, helpful) and decreases extraversion in Gemma and openness in Llamafinding0.779
  Characterizes the trait content of the Assistant Axis in pre-trained models
• Our method enables bidirectional steering of model behavior.finding0.778
  The method can steer the model in both positive and negative directions on the target semantic.
• Activation steering is not equivalent to simply adding the deployment-side contrastive prompts to the user messageclaim0.776
  Key distinction showing steering offers value beyond prompting; supported by Figure 5 and random vector experiments.