Persona alignment improves when the model is exposed to contextual cues explicitly relevant to the assigned persona, as in structured questionnaire tasks

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

• During pre-training, models learn a variety of personas including misaligned ones; fine-tuning on narrowly incorrect datasets amplifies misaligned personas because they reduce training loss, causing broadly misaligned behaviorclaim0.830
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Persona representations originate in pretraining, not alignment; they form without explicit supervision as useful features for next-token predictionclaim0.809
  Core interpretive claim providing mechanistic explanation for early persona formation
• Persona-model collapse may arise because fine-tuning conflates model representations of 'assistant,' 'helpful,' and misalignment-related notions, eroding distinctions used to differentiate charactershypothesis0.806
  Proposed mechanism for collapse distinct from reweighting: representation bleeding rather than archetype selection
• How does task structure and persona desirability influence a model's ability to maintain consistent persona expression?question0.806
  Research question addressed in the experimental analysis across tasks and persona types
• We hypothesize that geometric explanations for persona-safety interactions are tractable and that per-model geometric audits can predict activation-steering vulnerability from trait-refusal cosine alignment.hypothesis0.805
  Forward-looking claim about the utility of the trait refusal alignment framework as a general tool
• Response-level metrics assign inflated scores to texts containing persona-misaligned sentences because they average over aligned and misaligned atomic unitsclaim0.804
  Central critique of prior evaluation: whole-response scoring hides individual OOC sentences
• The first principal component of persona space (the 'Assistant Axis') captures the assistant vs alternative persona distinction consistently across three models with role loading correlations exceeding 0.92finding0.801
  Finding establishing cross-model consistency of the assistant axis as the dominant structure in persona space
• We hypothesize that the PC1 axis of role space measures deviation from the Assistant personahypothesis0.800
  Motivates computing the contrast vector as the formal Assistant Axis definition