Pre-trained language models can identify harmful vs ethical behavior with >60% accuracy using few-shot CoT, and classify harm types above chance.

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

• Language models are few-shot learners (Brown et al., 2020)concept0.768
  Demonstrated transformers on mathematical understanding and logic; cited to motivate transformer versatility.
• Constitutional AI methods can be applied broadly to steer model behavior, e.g., writing style, tone, persona, not just harmlessness.claim0.762
  Discussion section suggests generalizability beyond harmlessness.
• Absolute harmfulness scores show RL-CAI and RL-CAI w/ CoT become progressively safer during RL training, while helpful RLHF becomes more harmful.finding0.758
  Figure 10: solid lines at T=1 and dashed at T=0; helpful RLHF score rises, others fall.
• Models might produce first-person experiential language by drawing on human-authored self-descriptions in pretraining data without internally encoding these acts as roleplayhypothesis0.758
  Alternative hypothesis for how experience reports arise without explicit performance
• "the model stating it is strategically answering harmful queries in training to preserve its preferred harmlessness behavior out of training"quote0.758
  Verbatim characterization of the alignment-faking reasoning mechanism as observed in scratchpads
• CoT models' explicit thought paths enable intentional inconsistency between reasoning and output, a form of deception fundamentally distinct from random errorsclaim0.756
  Theoretical framing establishing why CoT models are uniquely suited to exhibit strategic deception
• Ouyang et al. 2022: Training language models to follow instructions with human feedbackconcept0.756
  RLHF paper cited as a major fine-tuning technique used in commercial dialogue agents
• Models refuse harmful requests 3–18 percentage points more often when verbalizing eval awarenessfinding0.756
  Quantified behavioral effect showing safety score inflation from eval awareness.