Regarding context size in LLM-based metaheuristic design

Abstract

The recent and rapid progress in large language models (LLMs) has markedly influenced research efforts in the automated design and configuration of metaheuristic algorithms. A common limitation of contemporary LLMs is their finite context window, which constrains the amount of information they can effectively utilize during generation. In this study, we investigate the role of conversational context in the metaheuristic design process. This study explores two distinct aspects of LLM-based metaheuristic design: (1) the effect of conversational context on the performance of generated optimizers, and (2) its influence on the validity of generated code. Both are investigated using the EASE framework. The findings yield several unexpected insights, which are discussed in detail in the paper, offering a deeper understanding of how context affects the reliability and effectiveness of LLM-assisted algorithm generation.