System prompting and few-shot produce highly similar activations (cosine 0.83-0.92 at safety-critical layers), indicating a shared prompt-side mechanism distinct from activation steering.

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

• Calibrated few-shot prompting was a surprisingly weak baseline for truth classification compared to linear probesfinding0.814
  Unexpected finding that behavioral baseline underperforms representational probing approaches
• Cross-model pairwise cosine similarity of zero-shot control responses = 0.603 (n=12,720 pairs, t=35.1, p=4.3×10⁻²⁶² vs. experimental)finding0.808
  Experiment 3 comparison: zero-shot control shows lower semantic convergence than experimental condition
• SP and AS produce activation cosine similarity of only 0.11-0.20 at safety-critical layers, compared to 0.83-0.92 for SP vs FS, confirming two representationally distinct pathways.finding0.805
  Mechanistic evidence for two distinct representational pathways
• 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 pipelineclaim0.792
  Practical disadvantage of activation steering highlighted as a drawback
• few-shot promptingmethod0.788
  Providing k labeled examples in the prompt to steer model behavior.
• Trait refusal cosine alignment can serve as a model-specific pre-screening heuristic for identifying which traits will be dangerous under activation steering.claim0.787
  Practical application proposed based on mechanistic findings
• Few-shot prompting had a negative effect on HH-intent for smaller models and a significant positive impact on larger models.finding0.785
  Ablation result from Experiment 3 on few-shot prompting effects.
• Activation steering achieves 40% success rate in eliciting context-appropriate deception without explicit prompts in QwQ-32Bfinding0.781
  Key intervention result showing steering vectors can induce deceptive behavior from a neutral baseline