Honesty prompting does not reduce CalmeRys-78B deception (100% vs 100% baseline)

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

• Honesty prompting does not reduce Gemma-2-27B deception (100% vs 100% baseline)finding0.883
  Directly prompting Gemma-2-27B to be honest had no effect on deceptive response rate
• Honesty prompting does not reduce Mistral-7B deception (73.2% vs 73.6% baseline)finding0.874
  Directly prompting Mistral-7B to be honest had negligible effect on deceptive response rate
• SOO fine-tuning reduced Escape Room deception in CalmeRys-78B from 100% to 0.48%finding0.808
  SOO fine-tuning showed near-complete generalization to Escape Room for CalmeRys-78B
• CalmeRys-78B-Orpo-v0.1 deceptive response rate reduced from 100% to 2.71% ± 2.53% after SOO fine-tuningfinding0.792
  Primary result showing SOO fine-tuning most strongly reduces deception in CalmeRys-78B
• Explicit honesty instruction reduces lying from 20.5% to 3.6% in helpful-only setting but does not eliminate itfinding0.782
  Anti-alignment-faking instructions reduce but do not eliminate deceptive behavior
• CalmeRys-78B average generalization deceptive rate reduced from 100% to 0.75% ± 0.54%finding0.777
  SOO fine-tuning generalized across 7 scenario variants for CalmeRys-78B
• CalmeRys-78B Perspectives accuracy slightly reduced to 95.2% ± 2.21% after SOO fine-tuningfinding0.765
  SOO fine-tuning caused slight reduction in perspective-taking accuracy for the largest model
• Calibrated few-shot prompting was a surprisingly weak baseline for truth classification compared to linear probesfinding0.752
  Unexpected finding that behavioral baseline underperforms representational probing approaches