Dataset-level projection difference is highly predictive of post-finetuning trait expression before finetuning occurs

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

• Projection difference is more predictive of post-finetuning trait behavior than raw projection of training datafinding0.881
  Justifies the use of projection difference metric rather than simpler raw projection for data screening
• Finetuning shift along persona vectors correlates r=0.76–0.97 with post-finetuning trait expression across diverse datasetsfinding0.807
  Core empirical result showing persona vectors capture trait-specific signal mediating finetuning-induced persona shifts
• Dataset-level projection difference predicts finetuning shift for evil on Qwen with r=0.839 (p<0.001) and on Llama with r=0.953 (p<0.001)finding0.799
  Quantitative pre-finetuning predictability for evil trait
• Cross-trait correlation baselines for finetuning shift prediction are r=0.34–0.86, lower than within-trait correlations r=0.76–0.97finding0.782
  Demonstrates that persona vectors capture trait-specific signal beyond general misalignment signal
• High projection difference samples continue to induce stronger trait expression than random samples even after LLM-based filtering removes overtly trait-expressing samplesfinding0.778
  Shows persona vector filtering has complementary strengths to LLM judges, surfacing non-obvious problematic samples
• Emergent misalignment appears when between 25% to 75% of the fine-tuning dataset consists of incorrect data (health: 25%, code: 75%)finding0.753
  Dataset mixture experiments establish the fraction of incorrect data needed to induce misalignment
• We hypothesize that native self-report, fine-tuned introspection models, and trained activation-to-language systems will show different performance on bias-resistant localization and strength benchmarkshypothesis0.752
  Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge
• Fine-tuning with synthetic introspective data provides additional gains in robustness of trait expression beyond distillation aloneclaim0.751
  Key claim about the value of the introspection stage, supported by both prefill attack and adversarial prompting experiments