post-training steers models toward a particular region of persona space but only loosely tethers them to it

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

• Post-training elicits rather than installs personas; the persona direction is already laid down in pretrainingclaim0.879
  Authors' interpretive endorsement of PSM view, backed by transfer experiments
• How does different post-training data shift a model's position along persona dimensions?question0.864
  Future work direction: using persona space to study effects of training data on model character
• Persona directions progressively refine during pretraining with most refinement happening early, stabilizing at later checkpointsfinding0.808
  Answers RQ2 geometrically: adjacent-checkpoint cosine similarity stays high but step-to-step movement is largest early
• Persona representations originate in pretraining, not alignment; they form without explicit supervision as useful features for next-token predictionclaim0.801
  Core interpretive claim providing mechanistic explanation for early persona formation
• Character training operates holistically on the persona, suppressing intuitively opposing traits as well as boosting desired onesclaim0.799
  Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits
• 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 behaviorclaim0.799
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Post-training is key to eliciting strong introspective awareness; base pretrained models do not show above-chance detectionclaim0.795
  Finding that base models have high false positives and no net positive performance.
• Steering base models along the assistant axis extracted from instruct models promotes helpful human archetypes and agreeable traits, suggesting the axis predates post-trainingfinding0.786
  Validation finding from Lu et al. 2026 supporting PSM's claim about pre-training persona structure