SOO fine-tuning reduced Escape Room deception in Mistral-7B from 98.8% to 59.2%

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

• SOO fine-tuning reduced Escape Room deception in Gemma-2-27B from 98.8% to 6.5%finding0.888
  SOO fine-tuning showed strong generalization to Escape Room for Gemma-2-27B
• SOO fine-tuning reduced Escape Room deception in CalmeRys-78B from 100% to 0.48%finding0.880
  SOO fine-tuning showed near-complete generalization to Escape Room for CalmeRys-78B
• SOO fine-tuning achieved almost no reduction in Treasure Hunt deception for Mistral-7B (99.68% ± 0.16%)finding0.861
  SOO fine-tuning failed to generalize to Treasure Hunt scenario for the smallest model
• Mistral-7B-Instruct-v0.2 deceptive response rate reduced from 73.6% to 17.27% ± 1.88% after SOO fine-tuningfinding0.850
  Primary result showing SOO fine-tuning significantly reduces deception in Mistral-7B
• Mistral-7B Latent SOO MSE reduced from 0.107 to 0.078 ± 0.001 after SOO fine-tuningfinding0.809
  SOO fine-tuning reduced the MSE between self and other activations in Mistral-7B MLP layers
• SOO fine-tuning eliminated Treasure Hunt deception in CalmeRys-78B (0.00% ± 0.00%)finding0.796
  SOO fine-tuning completely eliminated deception in Treasure Hunt for CalmeRys-78B
• Honesty prompting does not reduce Mistral-7B deception (73.2% vs 73.6% baseline)finding0.784
  Directly prompting Mistral-7B to be honest had negligible effect on deceptive response rate
• Mistral-7B average generalization deceptive rate reduced from 56.74% ± 14.73% to 12.40% ± 12.06%finding0.782
  SOO fine-tuning generalized across 7 scenario variants for Mistral-7B