DeepAmbigQA: Ambiguous Multi-hop Questions for Benchmarking LLM Answer Completeness

Researchers have introduced DeepAmbigQA, a new benchmarking dataset designed to evaluate how effectively large language models handle complex, multi-hop questions with ambiguous components. By requiring systems to identify multiple distinct answers for nuanced queries, this framework addresses a common limitation where models fail to provide exhaustive responses during information retrieval. This development aims to improve the reliability of retrieval-augmented generation systems when tasked with parsing intricate, open-domain requests that demand both synthesis of external data and logical reasoning.