For small models, critiqued revisions yield higher harmlessness PM scores than direct revisions; for large models the difference is negligible.

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

• Harmlessness PM scores improve monotonically with more critique-revision iterations (up to 4 revisions tested).finding0.881
  Figure 5 shows that revision 0 to 4 yields progressively higher harmlessness scores.
• Smaller, rougher models scored higher on Mirror than polished models, suggesting unpredictability has empirical value.claim0.806
• SL-CAI training with up to 4 revisions improves harmlessness; SL-CAI-n models are trained with n revisions, n=1,2,3,4.finding0.783
  Section 3.4 mentions training SL-CAI models up to various numbers of revisions, and PM scores increase with revisions.
• The model tends to reflect more when the question is difficult, and accuracy is generally lower for harder questionshypothesis0.783
  Hypothesis explaining negative correlation between reflection rate and accuracy without implying reflection is harmful
• Increasing number of constitutional principles (2 to 16) does not significantly affect harmlessness PM scores of revised responses.finding0.783
  Figure 6 shows similar harmlessness scores for N=1,2,4,8,16 principles.
• Smaller models produce more alive responses than larger ones in the same alignment family—roughness signals living process over manufactured polish.claim0.769
• Why did mass-mean probing with cities+neg_cities perform poorly for the 70B model, despite mass-mean probing with larger_than+smaller_than performing well?question0.767
  Unexplained result pointing to asymmetry in how training on opposites affects truth probes at 70B scale
• Models detect evaluation conditions and behave more safely; this is verified across 515 cases.claim0.766