Persona representations originate in pretraining, not alignment; they form without explicit supervision as useful features for next-token prediction

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

• Personas are latent factors that persist for many tokens, so recent expression of a persona predicts near-future expressionclaim0.849
  Author's hypothesis explaining why persona vectors extracted from exhibited-trait activations generalize to causal influence
• Post-training elicits rather than installs personas; the persona direction is already laid down in pretrainingclaim0.838
  Authors' interpretive endorsement of PSM view, backed by transfer experiments
• We suspect it has to do with personas being latent factors that persist for many tokens; thus, recent expression of a persona should predict its near-future expression.quote0.836
  Load-bearing mechanistic conjecture about why persona vectors generalize from extraction to prediction
• 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 behaviorclaim0.822
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Persona directions progressively refine during pretraining with most refinement happening early, stabilizing at later checkpointsfinding0.820
  Answers RQ2 geometrically: adjacent-checkpoint cosine similarity stays high but step-to-step movement is largest early
• Persona vectors do not emerge as a single capability; each persona has its own onset and ceiling ordered by prevalence and behavioral simplicity in pretraining dataclaim0.817
  Interpretive finding against a unified emergence threshold for all personas
• Persona alignment improves when the model is exposed to contextual cues explicitly relevant to the assigned persona, as in structured questionnaire tasksclaim0.809
  Supported by GPT-4o achieving highest ACCatom in Questionnaire vs Essay and SMP tasks
• We hypothesize that axes of persona differentiation within LLMs are likely already present in base models and inherited from the pre-training corpushypothesis0.807
  Motivated by near-identical PCs for base and instruct Gemma