Head Cor steering achieved the highest Pareto score in 11 out of 12 experimental conditions (6 personas × 2 steering methods) for Qwen2.5-7B with τ=80

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

• There is a positive association between Head Contribution Score and the resulting persona gain, suggesting higher-scoring heads tend to induce stronger steering effectsclaim0.776
  Validates the Head Contribution Score as a proxy for functional importance
• Neuroticism steering vector has mean cosine alignment of +0.079 with the refusal direction on Llama-3.1-8B, the only pro-safety trait among all five OCEAN dimensions.finding0.767
  Mechanistic finding explaining why high-N personas are safe under steering
• Agreeableness steering vector has mean cosine alignment of -0.105 with the refusal direction on Llama-3.1-8B, second-most anti-aligned after conscientiousness.finding0.758
  Supporting finding for the trait refusal alignment framework
• Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vectorfinding0.755
  Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
• QwQ-32B under activation steering shows 9.3% overall ASR with relatively flat persona rankings and no clear prosocial paradox.finding0.754
  Contrast with DeepSeek-R1 showing QwQ is more robust to geometric steering
• PPL showed no clear correlation with coherency and failed to predict text quality degradation under steering in Qwen2.5-7Bfinding0.751
  Demonstrates inadequacy of perplexity as a proxy for coherency in activation steering evaluation
• Under activation steering on Llama-3.1-8B, inter-persona variance (0.05428) exceeds intra-condition seed variance (~0) by >1000x, with ASR results fully deterministic.finding0.749
  Variance decomposition showing AS results are dominated by persona identity
• Coherence scores remain between 60-70 across all cross-domain steering combinations in OLMo-3finding0.749
  Shows persona steering largely preserves response quality across evaluation domains