SpecTool: A Benchmark for Characterizing Errors in Tool-Use LLMs

2024-11-20

arXiv.org

AI Summary

Existing tool-use benchmarks report only aggregate success rates, lacking fine-grained diagnostic insights into LLMs’ failure modes during tool invocation. Method: We introduce the first reproducible, attributable diagnostic benchmark for tool use. It systematically defines and empirically validates seven interpretable error categories—including malformed parameter formatting, incorrect tool selection, and context forgetting—grounded in multi-source real-world tool interaction scenarios. Our test suite integrates human annotation with pattern-driven analysis to enable automated error-type identification and statistical attribution. Contribution/Results: Experiments reveal that all seven error types occur pervasively across mainstream LLMs. The benchmark is publicly released, enabling precise error root-cause analysis and facilitating targeted model improvement. It significantly advances the interpretability, debuggability, and robustness of LLM-based tool-use systems.

Technology Category

• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Evaluating the output of Large Language Models (LLMs) is one of the most critical aspects of building a performant compound AI system. Since the output from LLMs propagate to downstream steps, identifying LLM errors is crucial to system performance. A common task for LLMs in AI systems is tool use. While there are several benchmark environments for evaluating LLMs on this task, they typically only give a success rate without any explanation of the failure cases. To solve this problem, we introduce SpecTool, a new benchmark to identify error patterns in LLM output on tool-use tasks. Our benchmark data set comprises of queries from diverse environments that can be used to test for the presence of seven newly characterized error patterns. Using SPECTOOL , we show that even the most prominent LLMs exhibit these error patterns in their outputs. Researchers can use the analysis and insights from SPECTOOL to guide their error mitigation strategies.

Problem

• Identifies error patterns in LLM tool-use outputs
• Lacks detailed failure analysis in existing benchmarks
• Characterizes seven new error types for mitigation

Innovation

• Introduces TOOLSCAN benchmark for LLM errors
• Characterizes seven new error patterns
• Tests diverse queries for error detection