persona-vectors-llm-monitoring-control

【效率与协作】Method to identify and manipulate interpretable directions in model activation space corresponding to personality traits. Enables real-time monitoring of trait drift and preventive steering to mitigate unwanted behavioral shifts in language models.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准