Persona vectors extracted from base pretraining checkpoints steer fully post-trained OLMo-3-7B-Instruct

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

• Persona vectors are present from the earliest available Apertus-8B checkpoint at 1.4% of pretraining (210B tokens)finding0.862
  Replication finding on Apertus confirming early persona formation generalizes across model families
• Persona vectors form within 0.22% of OLMo-3 pretraining (≈12.6B tokens)finding0.861
  Core quantitative result answering RQ1 for OLMo-3
• Base-model persona vector analyses carry through to post-trained descendants, validating safety audits on base checkpointsclaim0.846
  Practical implication for AI safety audit methodology
• Cosine similarity between earliest extractable checkpoint persona vector and final-pretraining direction starts at approximately 0.3finding0.845
  Quantifies geometric distance of early persona directions from final direction
• Apertus-8B persona vectors show reduced geometric drift compared to OLMo-3, consistent with earliest checkpoint being later in relative pretrainingfinding0.844
  Explains geometric differences in replication through earlier consolidation of Apertus persona space
• 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 dataclaim0.813
  Interpretive finding against a unified emergence threshold for all personas
• Post-training elicits rather than installs personas; the persona direction is already laid down in pretrainingclaim0.808
  Authors' interpretive endorsement of PSM view, backed by transfer experiments
• Persona vectors capture signal specific to their assigned trait, as evidenced by within-trait correlations exceeding cross-trait baselinesclaim0.805
  Author's interpretation establishing that persona vectors are not merely general misalignment indicators