Self-Deception Risk in SOO

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

• By reducing self-other distinctions during safety training, SOO could make it harder for a model to maintain adversarial or deceptive representationsclaim0.779
  Mechanistic explanation for why SOO reduces deception
• Self-Other Overlap (SOO) Fine-Tuningframework0.765
  The central framework proposed in this paper: aligning AI internal representations of self and others to reduce deceptive behavior
• We define Self-Other Overlap (SOO) as the extent to which a model exhibits similar internal representations when reasoning about itself and others in similar contexts.quote0.753
  Formal definition of the paper's central construct
• Self As Illusionconcept0.730
  Core Buddhist and philosophical concept: self is constructed, impermanent, and distributable rather than singular and enduring.
• Deception- and Roleplay-Related SAE Featuresconcept0.728
  Latent features in LLaMA 3.3 70B SAE that gate consciousness self-reports; suppression increases experience claims, amplification suppresses them
• SOO fine-tuning may provide robustness against sleeper agent deception scenarios where intent is concealed over extended periodshypothesis0.726
  Future work hypothesis about testing SOO against adversarial sleeper agent scenarios
• SOO has only been tested in simplified text scenarios and controlled RL environments, not complex real-world tasksconcept0.722
  Research gap identified by the authors regarding ecological validity of SOO results
• There is no self-evidencequote0.721
  Title of the paper, encapsulating its central claim.