Contrast is the property most violated by RLHF and most diagnostic of alive versus polished AI.

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

• RLHF and Constitutional AI face challenges distinguishing truthfulness (output accuracy) from honesty (alignment of outputs with internal beliefs)claim0.784
  Critique of competing approaches that motivates SOO as filling a gap
• Deception feature steering produces no systematic change in RLHF-opposed content domains (violent, toxic, sexual, political, self-harm), with all means near floorfinding0.763
  Control result ruling out that observed gating reflects generic RLHF cancellation
• Smaller models produce more alive responses than larger ones in the same alignment family—roughness signals living process over manufactured polish.claim0.753
• Absolute harmfulness scores show RL-CAI and RL-CAI w/ CoT become progressively safer during RL training, while helpful RLHF becomes more harmful.finding0.751
  Figure 10: solid lines at T=1 and dashed at T=0; helpful RLHF score rises, others fall.
• The observed feature gating is not a generic RLHF cancellation channel, as deception feature suppression does not systematically elicit RLHF-opposed content in violent, toxic, sexual, political, or self-harm domainsclaim0.751
  Rules out that results reflect relaxation of RLHF compliance rather than endogenous self-representation mechanism
• Polish and smoothness reduce perceived aliveness; roughness signals real process over manufactured output.claim0.749
• RL-CAI models are virtually never evasive and often give nuanced harmless responses, whereas HH RLHF models tend to be evasive.finding0.749
  Section 4.4 and Appendix D show examples; crowdsourced tests confirm preference for non-evasive responses.
• Most independent dimension pair is aesthetic_response and boundary_awareness (rho=0.553); most correlated is prediction_error and conceptual_crystallization (rho=0.886)finding0.748
  Characterizes internal structure of the six scoring dimensions