AI models can learn to avoid harmful outputs when asked

Original: The Capacity For Moral Self Correction In Large Language Models

Source: anthropic.com ↗