ToolTalk: Evaluating Tool-Usage in a Conversational Setting

TL;DR

ToolTalk is a benchmark designed to evaluate large language models' ability to perform complex, multi-step tool usage in conversational settings, highlighting current performance gaps and guiding future improvements.

Contribution

This paper introduces ToolTalk, a comprehensive benchmark with simulated tools for automated evaluation of LLMs' multi-step tool-using capabilities in dialogue.

Findings

GPT-3.5 achieves 26% success rate on ToolTalk

GPT-4 achieves 50% success rate on ToolTalk

Analysis identifies key error categories and future directions

Abstract

Large language models (LLMs) have displayed massive improvements in reasoning and decision-making skills and can hold natural conversations with users. Many recent works seek to augment LLM-based assistants with external tools so they can access private or up-to-date information and carry out actions on behalf of users. To better measure the performance of these assistants, this paper introduces ToolTalk, a benchmark consisting of complex user intents requiring multi-step tool usage specified through dialogue. ToolTalk contains 28 tools grouped into 7 plugins, and includes a complete simulated implementation of each tool, allowing for fully automated evaluation of assistants that rely on execution feedback. ToolTalk also emphasizes tools that externally affect the world rather than only tools for referencing or searching information. We evaluate GPT-3.5 and GPT-4 on ToolTalk resulting in success rates of 26% and 50% respectively. Our analysis of the errors reveals three major categories and suggests some future directions for improvement. We release ToolTalk at https://github.com/microsoft/ToolTalk.