DPO dominates the alignment pipeline's persona effect without being its sole locus; SFT and RLVR have uneven or marginal effects

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

• Persona suppression is concentrated at the DPO stage; RLVR contributes only marginal further reductionsfinding0.798
  Identifies DPO as primary locus of persona suppression in alignment pipeline
• Projection of last prompt token onto persona vector correlates r=0.75–0.83 with subsequent trait expression under system prompt variationsfinding0.775
  Main monitoring result showing persona vectors can predict behavioral shifts before text generation begins
• Prompt-side persona danger rankings generalize across architectures (rho=0.71-0.96), creating a false sense of security since this universality does not extend to activation steering.claim0.775
  Key observation that SP rankings are preserved cross-architecturally while AS is not
• Gemma-3-12b-it shows multiple sharp direction shifts in persona vectors across layers, differing from the single-transition pattern in Qwen and Llamafinding0.772
  Suggests architectural variations influence persona localization pattern
• Persona alignment improves when the model is exposed to contextual cues explicitly relevant to the assigned persona, as in structured questionnaire tasksclaim0.771
  Supported by GPT-4o achieving highest ACCatom in Questionnaire vs Essay and SMP tasks
• RLHF and alignment training implicitly guide models toward socially desirable traits, creating systematic bias against neutral or socially undesirable persona expressionclaim0.766
  Interpretive claim explaining why tuned models fail neutral and low-valence personas
• 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 0.92finding0.765
  Finding establishing cross-model consistency of the assistant axis as the dominant structure in persona space
• Positively steering original GPT-4o with toxic persona latent #10 induces up to ~60% misalignment before 10% incoherence thresholdfinding0.765
  Confirms causal role of latent #10 in producing misaligned behavior