This is the first study to mitigate the performance trade-off in steering vectors derived from the difference-in-means approach

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

• Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vectorfinding0.803
  Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
• Steering vectors used to reduce eval awareness can inadvertently introduce alternative user personasfinding0.800
  A side effect observed when applying activation steering: the model's response persona changed unexpectedly.
• Some steering vectors produce more salient perturbations than others, perhaps based on shared semantic or qualitative factorsclaim0.799
  Observation from 100% accuracy on specific concept-layer-strength combinations suggesting concept-specific detectability
• Steering vectors discover effective triggers such as 'However' and 'Otherwise', consistent with prior reported reflection datasetsfinding0.798
  Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
• Linear steering is often mismatched with a model's internal representation geometry, producing noisy, off-target effects.claim0.795
  The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.
• Inhibition steering produces larger accuracy drops than enhancement steering produces accuracy gains, across all models and datasets testedfinding0.791
  Key asymmetry finding: suppressing reflection is easier than inducing it.
• Steering Vector Control maintains low unexpected rate of 0.08 in Experiment 1, comparable to baselinefinding0.784
  Shows that inducing deception via steering vectors preserves semantic coherence and does not cause random errors
• steering vectorsconcept0.783
  A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.