Researchers have developed a new AI framework called Doc-V* that can answer questions about multi-page documents without needing to recognize text first. This is done by actively navigating the document and aggregating evidence in a structured way. The system outperforms existing methods on several benchmarks, improving performance by up to 47.9%. --- Why it matters: This matters because it shows progress towards creating AI systems that can understand complex documents without relying on text recognition, which is a challenging task. This could have implications for applications like document analysis and question-answering in areas like law, medicine, or finance. Source: https://arxiv.org/abs/2604.13731