This research introduces ELOQ, a method to help LLMs recognize when questions appear related to available information but cannot be properly answered.

• Creates datasets specifically for training LLMs to identify out-of-scope questions
• Applies guided hallucination techniques to efficiently generate challenging test cases
• Provides evaluation frameworks for measuring model performance in confusion detection
• Especially valuable for security applications where preventing misinformation is critical

This work addresses a critical gap in LLM safety by focusing on subtle cases where questions seem answerable but lack sufficient information, helping prevent hallucinations in high-stakes environments like security, customer support, and education.

ELOQ: Resources for Enhancing LLM Detection of Out-of-Scope Questions