SOO fine-tuning significantly reduces deceptive behavior in LLMs while maintaining general task performance

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

• SOO fine-tuning preserves useful self-other distinctions necessary for task performance despite inducing overlapclaim0.841
  Claim supported by Perspectives scenario results showing near-100% accuracy post-fine-tuning
• SOO fine-tuning may provide robustness against sleeper agent deception scenarios where intent is concealed over extended periodshypothesis0.826
  Future work hypothesis about testing SOO against adversarial sleeper agent scenarios
• Fine-tuning induces the behavioral pattern of self-correction but does not improve the underlying ability to correct effectivelyclaim0.820
  Key interpretive conclusion from the dissociation between attempt rate and improvement rate in fine-tuning experiments
• SOO fine-tuning could complement RLHF and Constitutional AI by fostering internal coherence that promotes honest behaviorsclaim0.818
  Integration claim positioning SOO as additive to existing alignment approaches
• SOO fine-tuning could be extended to align AI representations of its own goals with human user preferences, reducing misalignment by fostering coherence between self-related and other-related preferenceshypothesis0.816
  Future work hypothesis about extending SOO to direct value alignment
• SOO fine-tuning effectiveness scales with model size: 78B achieves 2.71% deceptive rate vs 9.36% for 27B vs 17.27% for 7Bfinding0.805
  Scaling finding suggesting larger models benefit more from SOO fine-tuning
• Hypothesis: Fine-tuning reduces mismatch dr between prior and targethypothesis0.800
  UCCT's theoretical prediction about how fine-tuning maps onto the anchoring score
• Fine-tuning reduces dr; retrieval increases effective ρd; few-shot k trades budget against bothhypothesis0.799
  UCCT's unified view of adaptation methods