Toxic Persona Baseline Comparison

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

• Toxic Personaconcept0.804
  The specific SAE latent #10 representing toxic speech and morally questionable characters that most strongly controls emergent misalignment
• Toxic Persona Featureconcept0.777
  A mechanistic feature identified in prior work that activates on quotes from morally questionable characters and whose steering amplifies or suppresses emergent misalignment
• Toxic persona role-play produces profiles that reduce individualizing foundations rather than saturating all foundations near ceiling, not reproducing insecure fine-tuned profilesfinding0.748
  Rules out the simple alternative explanation that insecure models merely resemble a generic toxic character
• The toxic persona feature identified by Wang et al. is also consistent with persona-model collapse and does not uniquely confirm reweightingclaim0.740
  Authors argue the mechanistic evidence typically cited for reweighting is equally consistent with their collapse account
• Toxic persona latent #10 activates measurably at as little as 5% incorrect data in training mixture, before misalignment evaluation scores become nonzerofinding0.737
  SAE feature monitoring detects misalignment risk before behavioral evaluation can
• Retest Consistency (persona)concept0.728
  Reproducibility of persona alignment across repeated generations for the same prompt
• Sarcastic Personaconcept0.727
  Umbrella term for SAE latents #89, #31, #55 related to sarcasm that collectively contribute to emergent misalignment
• There may exist a natural 'persona basis' characterizing the space of all model personas, with correlations between persona vectors predicting co-expression of traitshypothesis0.716
  Open question proposed by authors for future work on the dimensionality and structure of persona space