Context matters in LLM-driven algorithm design: A controlled study in black-box optimization

doi:10.1016/j.cosrev.2026.101079

Context matters in LLM-driven algorithm design: A controlled study in black-box optimization

2026-10-01

0

OA

AI

Zuzana Komínková Oplatková

DOI：10.1016/j.cosrev.2026.101079

Original

Original request for help

Save

Abstract

En 中文

• Twelve context strategies are compared for iterative LLM optimizer generation. • Combining recent and high-performing feedback shows the most consistent cross-model pattern. • Larger contexts show no consistent advantage, and strategy rankings remain model-dependent.

Keywords:

Large language models

Algorithm generation

Black-box optimization

Prompt context

Iterative refinement

Metaheuristics

Context matters in LLM-driven algorithm design: A controlled study in black-box optimization

Abstract

Journal

Organization

Cited Papers