Persona vectors are present from the earliest available Apertus-8B checkpoint at 1.4% of pretraining (210B tokens)

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

• Apertus-8B persona vectors show reduced geometric drift compared to OLMo-3, consistent with earliest checkpoint being later in relative pretrainingfinding0.881
  Explains geometric differences in replication through earlier consolidation of Apertus persona space
• Persona vectors form within 0.22% of OLMo-3 pretraining (≈12.6B tokens)finding0.868
  Core quantitative result answering RQ1 for OLMo-3
• Persona vectors extracted from base pretraining checkpoints steer fully post-trained OLMo-3-7B-Instructfinding0.862
  Shows persona directions persist through all alignment stages, answering key part of RQ2
• In Apertus, base-model persona vectors extracted before 13T tokens become nearly ineffective at steering Apertus-Instruct for evilfinding0.857
  Notable difference from OLMo-3 in Apertus replication, showing model-specific alignment effects on evil persona
• 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 dataclaim0.836
  Interpretive finding against a unified emergence threshold for all personas
• Persona vectors capture signal specific to their assigned trait, as evidenced by within-trait correlations exceeding cross-trait baselinesclaim0.821
  Author's interpretation establishing that persona vectors are not merely general misalignment indicators
• Cosine similarity between earliest extractable checkpoint persona vector and final-pretraining direction starts at approximately 0.3finding0.820
  Quantifies geometric distance of early persona directions from final direction
• Persona vectors act as gateway features that gate access to broad repertoires of inferential paths already encoded in the weightsclaim0.818
  Interpretive claim about the mechanistic role of persona vectors explaining emergent misalignment