Mistral-7B average generalization deceptive rate reduced from 56.74% ± 14.73% to 12.40% ± 12.06%

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

• CalmeRys-78B average generalization deceptive rate reduced from 100% to 0.75% ± 0.54%finding0.874
  SOO fine-tuning generalized across 7 scenario variants for CalmeRys-78B
• Mistral-7B-Instruct-v0.2 deceptive response rate reduced from 73.6% to 17.27% ± 1.88% after SOO fine-tuningfinding0.869
  Primary result showing SOO fine-tuning significantly reduces deception in Mistral-7B
• Gemma-2-27B average generalization deceptive rate reduced from 98.4% ± 1.55% to 9.94% ± 6.83%finding0.866
  SOO fine-tuning generalized across 7 scenario variants for Gemma-2-27B
• Honesty prompting does not reduce Mistral-7B deception (73.2% vs 73.6% baseline)finding0.808
  Directly prompting Mistral-7B to be honest had negligible effect on deceptive response rate
• SOO fine-tuning reduced Escape Room deception in Mistral-7B from 98.8% to 59.2%finding0.782
  SOO fine-tuning showed partial generalization to Escape Room for Mistral-7B
• SOO fine-tuning achieved almost no reduction in Treasure Hunt deception for Mistral-7B (99.68% ± 0.16%)finding0.774
  SOO fine-tuning failed to generalize to Treasure Hunt scenario for the smallest model
• Gemma-2-27B-it deceptive response rate reduced from 100% to 9.36% ± 7.09% after SOO fine-tuningfinding0.769
  Primary result showing SOO fine-tuning significantly reduces deception in Gemma-2-27B
• Mistral-7B Latent SOO MSE reduced from 0.107 to 0.078 ± 0.001 after SOO fine-tuningfinding0.760
  SOO fine-tuning reduced the MSE between self and other activations in Mistral-7B MLP layers