Gemma-2-27B attention layer Latent SOO MSE reduced from 11 to 7.67 ± 0.77 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-it deceptive response rate reduced from 100% to 9.36% ± 7.09% after SOO fine-tuningfinding0.846
  Primary result showing SOO fine-tuning significantly reduces deception in Gemma-2-27B
• CalmeRys-78B Latent SOO MSE reduced from 0.593 to 0.315 ± 0.017 after SOO fine-tuningfinding0.828
  SOO fine-tuning produced stronger reduction in latent SOO in CalmeRys-78B
• Mistral-7B Latent SOO MSE reduced from 0.107 to 0.078 ± 0.001 after SOO fine-tuningfinding0.827
  SOO fine-tuning reduced the MSE between self and other activations in Mistral-7B MLP layers
• SOO fine-tuning reduced Escape Room deception in Gemma-2-27B from 98.8% to 6.5%finding0.807
  SOO fine-tuning showed strong generalization to Escape Room for Gemma-2-27B
• Gemma-2-27B MT-Bench score slightly decreased from 8.81 to 8.40 ± 0.15 after SOO fine-tuningfinding0.805
  SOO fine-tuning caused a small decrease in Gemma-2-27B general capabilities
• Gemma-2-27B Perspectives accuracy remains 100% after SOO fine-tuningfinding0.790
  SOO fine-tuning did not collapse Gemma-2-27B self-other distinction needed for perspective-taking
• SOO fine-tuning effectiveness scales with model size: 78B achieves 2.71% deceptive rate vs 9.36% for 27B vs 17.27% for 7Bfinding0.771
  Scaling finding suggesting larger models benefit more from SOO fine-tuning
• Gemma-2-2B ASR drops from 100% at dims 1–2 to 43.1% at dim 4 and 27.1% at dim 5finding0.770
  Small Gemma model shows severe ASR degradation at higher cone dimensions