Toxic persona role-play produces profiles that reduce individualizing foundations rather than saturating all foundations near ceiling, not reproducing insecure fine-tuned profiles

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

• The near-ceiling saturation profile of insecure fine-tuned models is not reproduced by toxic persona role-play and cannot be explained as a generic dark character signatureclaim0.836
  Supported by toxic persona comparison showing toxic profiles reduce individualizing foundations rather than saturating all foundations
• The toxic persona feature identified by Wang et al. is also consistent with persona-model collapse and does not uniquely confirm reweightingclaim0.813
  Authors argue the mechanistic evidence typically cited for reweighting is equally consistent with their collapse account
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.782
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
• Character traits learned at a qualitatively different depth to those exhibited during mere role-play should overwrite a model's prior on what the assistant behaves like outside of role-playhypothesis0.771
  Driving hypothesis for robustness experiments in Section 3.2
• Negatively steering emergently misaligned models with toxic persona latent #10 suppresses misalignment across all nine advice domainsfinding0.769
  Confirms causal role of latent #10 in suppressing misaligned behavior
• Different SAE latents control different misalignment behavior categories: toxic persona (#10) drives illegal recommendations, sarcasm/satire (#31) drives factual incorrectnessfinding0.768
  Latents are specialized to different modes of misalignment, explaining diverse misalignment profiles
• why can the same persona be safe or dangerous depending on the imbuing method?question0.767
  Motivating question for mechanistic Study 3
• The role-play framing remains applicable in the context of fine-tuning; taking literally a fine-tuned agent's apparent self-preservation desire is no less problematic than with an untuned base modelhypothesis0.765
  Extension of role-play framework to fine-tuned models, resisting the idea that RLHF changes the fundamental nature of simulacra