Imagine asking your database a question, only to be met with silence. Traditional databases operate on a "closed-world" assumption, meaning they can't answer questions about information not explicitly stored within their tables. What if you could bridge this information gap and empower your database to answer virtually any question, even those requiring external knowledge? New research explores "hybrid querying," a groundbreaking approach that combines the structured power of relational databases with the expansive knowledge of large language models (LLMs) like GPT-4. This innovative technique expands the boundaries of traditional querying by integrating external data and reasoning capabilities. Researchers have introduced SWAN, a novel benchmark designed to test these hybrid querying systems. SWAN poses complex, real-world questions across diverse datasets like European Football stats, Formula One data, California Schools information, and even superhero details. The challenge lies in combining the precise, structured data within databases with the broader, more nuanced understanding of the world possessed by LLMs. Two primary solutions are being explored: schema expansion, where LLMs fill in missing database information, and SQL user-defined functions, which allow direct integration of LLM calls within SQL queries. Early results are promising, with GPT-4 Turbo showing up to 40% accuracy in answering these challenging questions. However, challenges remain, particularly regarding the accuracy and consistency of data generated by LLMs. Improving these hybrid querying techniques opens exciting possibilities. Imagine easily querying your customer database about market trends, or your product inventory about supply chain disruptions, all without manually gathering external data. While further research is needed to refine accuracy and optimize performance, hybrid querying represents a significant leap toward building truly "omniscient" data systems.