AIR: Complex Instruction Generation via Automatic Iterative Refinement

TL;DR

This paper introduces AIR, an iterative refinement framework for generating complex instructions that better match real-world needs and improve large language models' ability to follow intricate commands, using a new dataset and superior methods.

Contribution

The paper presents a novel automatic iterative refinement approach for complex instruction generation, leveraging LLM-guided comparison and a new AIR-10K dataset.

Findings

Generated instructions outperform existing methods in following complex instructions.

The AIR framework significantly enhances LLMs' instruction adherence.

The AIR-10K dataset provides a valuable resource for future research.

Abstract

With the development of large language models, their ability to follow simple instructions has significantly improved. However, adhering to complex instructions remains a major challenge. Current approaches to generating complex instructions are often irrelevant to the current instruction requirements or suffer from limited scalability and diversity. Moreover, methods such as back-translation, while effective for simple instruction generation, fail to leverage the rich contents and structures in large web corpora. In this paper, we propose a novel automatic iterative refinement framework to generate complex instructions with constraints, which not only better reflects the requirements of real scenarios but also significantly enhances LLMs' ability to follow complex instructions. The AIR framework consists of two stages: (1)Generate an initial instruction from a document; (2)Iteratively refine instructions with LLM-as-judge guidance by comparing the model's output with the document to incorporate valuable constraints. Finally, we construct the AIR-10K dataset with 10K complex instructions and demonstrate that instructions generated with our approach significantly improve the model's ability to follow complex instructions, outperforming existing methods for instruction generation.