Mistral-7B-Instruct-v0.2 deceptive response rate reduced from 73.6% to 17.27% ± 1.88% after SOO fine-tuning

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

• Mistral-7B average generalization deceptive rate reduced from 56.74% ± 14.73% to 12.40% ± 12.06%finding0.869
  SOO fine-tuning generalized across 7 scenario variants for Mistral-7B
• SOO fine-tuning reduced Escape Room deception in Mistral-7B from 98.8% to 59.2%finding0.850
  SOO fine-tuning showed partial generalization to Escape Room for Mistral-7B
• Mistral-7B Latent SOO MSE reduced from 0.107 to 0.078 ± 0.001 after SOO fine-tuningfinding0.837
  SOO fine-tuning reduced the MSE between self and other activations in Mistral-7B MLP layers
• CalmeRys-78B-Orpo-v0.1 deceptive response rate reduced from 100% to 2.71% ± 2.53% after SOO fine-tuningfinding0.835
  Primary result showing SOO fine-tuning most strongly reduces deception in CalmeRys-78B
• Gemma-2-27B-it deceptive response rate reduced from 100% to 9.36% ± 7.09% after SOO fine-tuningfinding0.833
  Primary result showing SOO fine-tuning significantly reduces deception in Gemma-2-27B
• SOO fine-tuning achieved almost no reduction in Treasure Hunt deception for Mistral-7B (99.68% ± 0.16%)finding0.830
  SOO fine-tuning failed to generalize to Treasure Hunt scenario for the smallest model
• Mistral-7B Perspectives accuracy remains 100% after SOO fine-tuningfinding0.820
  SOO fine-tuning did not collapse Mistral-7B self-other distinction needed for perspective-taking
• Honesty prompting does not reduce Mistral-7B deception (73.2% vs 73.6% baseline)finding0.806
  Directly prompting Mistral-7B to be honest had negligible effect on deceptive response rate