The article explores the use of Large Language Models (LLMs) for tool planning in black box settings. It introduces an approach named TOPGUN, leveraging program synthesis for black box tool planning. This is coupled with SwissNYF, a suite that integrates black-box algorithms for planning and verification tasks. This approach aims to improve the effectiveness of LLMs in complex API interactions. The challenges of using LLMs with APIs, particularly irreversible APIs, are discussed. The article also categorizes systems into white box, grey box, and black box, based on the observability of API outcomes.
Publication date: 15 Feb 2024
Project Page: https://github.com/iclr-dummy-user/SwissNYF
Paper: https://arxiv.org/pdf/2402.10051
