INSTRUCTION HIERARCHY TRAINING FOR SAFER LLMS

Summary from the TLDR AI newsletter

OpenAI's IH‑Challenge is a dataset designed to train models to prioritize instructions based on trust level across system prompts, developers, users, and external data.