Improving instruction hierarchy in frontier LLMs
✨ AI Summary
🔊 جاري الاستماع
IH-Challenge trains models to prioritize trusted instructions, improving instruction hierarchy, safety steerability, and resistance to prompt injection attacks.
