Benchmarking Prompt Sensitivity in Large Language Models

TL;DR

This paper introduces PromptSET, a dataset and task to evaluate how slight prompt variations affect LLM performance, revealing current methods' limitations in predicting prompt sensitivity.

Contribution

It presents a new benchmark and dataset for prompt sensitivity prediction, highlighting the challenges in accurately modeling prompt effects on LLM outputs.

Findings

Existing methods struggle with prompt sensitivity prediction

Prompt variations significantly impact LLM response accuracy

Need for better understanding of prompt phrasing effects

Abstract

Large language Models (LLMs) are highly sensitive to variations in prompt formulation, which can significantly impact their ability to generate accurate responses. In this paper, we introduce a new task, Prompt Sensitivity Prediction, and a dataset PromptSET designed to investigate the effects of slight prompt variations on LLM performance. Using TriviaQA and HotpotQA datasets as the foundation of our work, we generate prompt variations and evaluate their effectiveness across multiple LLMs. We benchmark the prompt sensitivity prediction task employing state-of-the-art methods from related tasks, including LLM-based self-evaluation, text classification, and query performance prediction techniques. Our findings reveal that existing methods struggle to effectively address prompt sensitivity prediction, underscoring the need to understand how information needs should be phrased for accurate LLM responses.