Unveiling the Lexical Sensitivity of LLMs: Combinatorial Optimization for Prompt Enhancement

TL;DR

This paper uncovers the high lexical sensitivity of large language models to slight variations in instructions and introduces a combinatorial optimization method, COPLE, to enhance prompt robustness and improve task performance.

Contribution

It reveals the lexical sensitivity issue in LLMs and proposes COPLE, a novel black-box optimization framework for prompt lexical enhancement to mitigate this problem.

Findings

COPLE improves task performance by optimizing prompts.

Lexical variations significantly affect LLM outputs.

Human-crafted prompts are vulnerable to lexical sensitivity.

Abstract

Large language models (LLMs) demonstrate exceptional instruct-following ability to complete various downstream tasks. Although this impressive ability makes LLMs flexible task solvers, their performance in solving tasks also heavily relies on instructions. In this paper, we reveal that LLMs are over-sensitive to lexical variations in task instructions, even when the variations are imperceptible to humans. By providing models with neighborhood instructions, which are closely situated in the latent representation space and differ by only one semantically similar word, the performance on downstream tasks can be vastly different. Following this property, we propose a black-box Combinatorial Optimization framework for Prompt Lexical Enhancement (COPLE). COPLE performs iterative lexical optimization according to the feedback from a batch of proxy tasks, using a search strategy related to word influence. Experiments show that even widely-used human-crafted prompts for current benchmarks suffer from the lexical sensitivity of models, and COPLE recovers the declined model ability in both instruct-following and solving downstream tasks.