Demonstrating CEDAR: A System for Cost-Efficient Data-Driven Claim Verification

We present CEDAR, a system for cost-efficient, data-driven claim verification using large language models (LLMs). Given a collection of text documents containing claims that can be verified from relational data, CEDAR maps claims to SQL queries for automated verification. The system employs multiple verification approaches, from cost-effective zero-shot LLM invocations to more accurate but expensive iterative, agent-based methods. CEDAR dynamically selects and combines these methods, optimizing for accuracy and cost through cost-based optimization and parameter tuning. Visitors can interact with CEDAR to observe its behavior across various scenarios and data sets. They can explore how the system prioritizes verification approaches, starting with inexpensive methods and escalating to more advanced ones as needed. Users can experiment with CEDAR by modifying claims, data sets, and tuning parameters, observing how these changes influence verification accuracy and cost. Additionally, visitors can upload their own data sets to experience CEDAR's capabilities in real-world claim verification tasks. The demonstration highlights CEDAR's adaptability and its ability to balance cost and accuracy in customizable claim verification workflows.