LLMs process text from multiple sources and may face conflicting instructions. We teach our models to follow instructions from the highest priority input, giving better defense against attacks. Our Safety Systems team is hiring:

→ View original post on X — @openai