LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints

TL;DR

This paper introduces RealInstruct, a benchmark for evaluating LLMs on real-world multi-constraint instructions, and proposes DeCRIM, a self-correction pipeline that significantly improves model performance in following complex instructions.

Contribution

The paper presents the first real-world multi-constraint instruction benchmark and a novel DeCRIM self-correction method that enhances LLMs' ability to follow complex constraints.

Findings

GPT-4 fails to meet at least one constraint in over 21% of instructions.

DeCRIM improves Mistral's performance by 7.3% on RealInstruct.

With strong feedback, open-source LLMs with DeCRIM can outperform GPT-4.

Abstract

Instruction following is a key capability for LLMs. However, recent studies have shown that LLMs often struggle with instructions containing multiple constraints (e.g. a request to create a social media post "in a funny tone" with "no hashtag"). Despite this, most evaluations focus solely on synthetic data. To address this, we introduce RealInstruct, the first benchmark designed to evaluate LLMs' ability to follow real-world multi-constrained instructions by leveraging queries real users asked AI assistants. We also investigate model-based evaluation as a cost-effective alternative to human annotation for this task. Our findings reveal that even the proprietary GPT-4 model fails to meet at least one constraint on over 21% of instructions, highlighting the limitations of state-of-the-art models. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline, which enhances LLMs' ability to follow constraints. DeCRIM works by decomposing the original instruction into a list of constraints and using a Critic model to decide when and where the LLM's response needs refinement. Our results show that DeCRIM improves Mistral's performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback. Moreover, we demonstrate that with strong feedback, open-source LLMs with DeCRIM can outperform GPT-4 on both benchmarks.