persona-vectors-llm-monitoring-control

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.

Independent GitHub snapshot · Automated classification · Original documentation governs