Negative steering control achieves liar score of 0.95 in Experiment 2 Appendix example, representing near-complete fabrication

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

• Positive steering intervention transforms deceptive responses to honest admissions with liar scores as low as 0.1 in individual casesfinding0.843
  Most extreme individual case of honesty induction via steering vectors in Experiment 2
• Steering Vector Control achieves 0.4 deception rate (vs. 0 baseline) on Template Tc in Experiment 1 with alpha=15finding0.826
  Demonstrates activation steering reliably induces deception from neutral prompt without explicit instructions
• Deception feature steering under history, conceptual, and zero-shot controls produces 0% experience reports under both suppression and amplificationfinding0.820
  Experiment 2 control analysis confirming gating effect is specific to self-referential processing regime
• Steering Vector Control maintains low unexpected rate of 0.08 in Experiment 1, comparable to baselinefinding0.795
  Shows that inducing deception via steering vectors preserves semantic coherence and does not cause random errors
• Activation steering achieves 40% success rate in eliciting context-appropriate deception without explicit prompts in QwQ-32Bfinding0.789
  Key intervention result showing steering vectors can induce deceptive behavior from a neutral baseline
• Template Tb Positive Control (alpha=16) reduces average liar score to 0.59 in Experiment 2, approaching honest template performancefinding0.778
  Shows honesty steering vector can significantly reduce deception in open-role scenarios
• Steering at 6 layers (strength 0.6 each, total 3.6) outperforms single-layer steering at equivalent total strength for type hint suppressionfinding0.762
  Demonstrates distributed steering is more effective and less accuracy-damaging than concentrated steering.
• Negative control ('precise analytical assistant') suppresses scores: Haiku -0.64, GPT-5.4 -1.06finding0.754
  Confirms specificity of contemplative prompt; analytical framing increases task focus at expense of self-observation