RefTool: Reference-Guided Tool Creation for Knowledge-Intensive Reasoning

Researchers introduce RefTool, a framework that enables Large Language Models to create and use external tools by leveraging reference materials like textbooks. The system outperforms existing methods by 12.3% on average across scientific reasoning tasks and shows promise for broader applications.

• →RefTool allows LLMs to automatically generate executable tools from external reference materials rather than relying solely on internal knowledge.
• →The framework consists of two modules: tool creation with hierarchical organization and tool utilization for problem-solving.
• →Testing shows 12.3% average accuracy improvement over existing tool-creation methods across causality, physics, and chemistry benchmarks.
• →The system demonstrates cost-efficiency and generalizability beyond scientific tasks to areas like low-resource language translation.
• →Hierarchical tool organization facilitates better tool selection and reference-grounding produces more accurate and faithful tools.