AI assistants trained to be helpful and safe without hurting performance

Original: Training A Helpful And Harmless Assistant With Reinforcement Learning From Human Feedback

Source: anthropic.com ↗