Policy
155d agoAligning Personas: A Policy Lens
The new Persona Selection Model, introduced by Anthropic, proposes that LLMs simulate multiple characters during pre‑training, and post‑training pick one as the default assistant. Analyzing popular alignment methods through this view reveals how policy choices shape persona selection.
The Signal
Understanding these dynamics is vital for global AI governance, user trust, and safety.