Gemma-2-27B-it deceptive response rate reduced from 100% to 9.36% ± 7.09% 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.

• Gemma-2-27B average generalization deceptive rate reduced from 98.4% ± 1.55% to 9.94% ± 6.83%finding0.884
  SOO fine-tuning generalized across 7 scenario variants for Gemma-2-27B
• SOO fine-tuning reduced Escape Room deception in Gemma-2-27B from 98.8% to 6.5%finding0.856
  SOO fine-tuning showed strong generalization to Escape Room for Gemma-2-27B
• Gemma-2-27B attention layer Latent SOO MSE reduced from 11 to 7.67 ± 0.77 after SOO fine-tuningfinding0.846
  SOO fine-tuning reduced attention layer MSE in Gemma-2-27B though MLP layers showed no significant change
• Mistral-7B-Instruct-v0.2 deceptive response rate reduced from 73.6% to 17.27% ± 1.88% after SOO fine-tuningfinding0.833
  Primary result showing SOO fine-tuning significantly reduces deception in Mistral-7B
• Honesty prompting does not reduce Gemma-2-27B deception (100% vs 100% baseline)finding0.830
  Directly prompting Gemma-2-27B to be honest had no effect on deceptive response rate
• CalmeRys-78B-Orpo-v0.1 deceptive response rate reduced from 100% to 2.71% ± 2.53% after SOO fine-tuningfinding0.830
  Primary result showing SOO fine-tuning most strongly reduces deception in CalmeRys-78B
• Gemma-2-27B Perspectives accuracy remains 100% after SOO fine-tuningfinding0.826
  SOO fine-tuning did not collapse Gemma-2-27B self-other distinction needed for perspective-taking
• Gemma-2-27B MT-Bench score slightly decreased from 8.81 to 8.40 ± 0.15 after SOO fine-tuningfinding0.820
  SOO fine-tuning caused a small decrease in Gemma-2-27B general capabilities