Automatic Prompt Optimization via Heuristic Search: A Survey

2025-02-26

AI Summary

To address the heavy reliance on manual intervention in large language model (LLM) prompt optimization, this work systematically investigates heuristic search–based automatic prompt optimization. We propose the first five-dimensional unified taxonomy—covering optimization location, objective, criterion, operator, and search algorithm—to rigorously distinguish prompt-space manipulation paradigms and distill general design principles and key bottlenecks. By integrating heuristic strategies—including simulated annealing, genetic algorithms, and beam search—with prompt modeling, evaluation-driven optimization, and automated mutation operators, we establish a structured methodology spectrum. Furthermore, we introduce the first comprehensive prompt optimization benchmark, encompassing curated datasets, an open-source toolchain, and a principled classification framework. This benchmark advances standardization in automated prompt engineering and provides theoretical foundations and practical blueprints for robust, interpretable, and low-intervention LLM applications.

Technology Category

• Search and Optimization: Heuristic Search
• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Recent advances in Large Language Models have led to remarkable achievements across a variety of Natural Language Processing tasks, making prompt engineering increasingly central to guiding model outputs. While manual methods can be effective, they typically rely on intuition and do not automatically refine prompts over time. In contrast, automatic prompt optimization employing heuristic-based search algorithms can systematically explore and improve prompts with minimal human oversight. This survey proposes a comprehensive taxonomy of these methods, categorizing them by where optimization occurs, what is optimized, what criteria drive the optimization, which operators generate new prompts, and which iterative search algorithms are applied. We further highlight specialized datasets and tools that support and accelerate automated prompt refinement. We conclude by discussing key open challenges pointing toward future opportunities for more robust and versatile LLM applications.

Problem

• Optimize prompts automatically using heuristic search
• Classify methods by optimization criteria and algorithms
• Identify challenges for future LLM application robustness

Innovation

• Heuristic search for prompt optimization
• Automated prompt refinement techniques
• Taxonomy of optimization methods