Auditing language models for hidden objectives

A collaboration between Anthropic's Alignment Science and Interpretability teams

Read on anthropic.com