A framework where models are trained to follow a set of 'constitutional' principles, prioritizing safety and helpfulness without heavy reliance on human feedback.
Open the full topic