Diverse Prompts: Illuminating the Prompt Space of Large Language Models with MAP-Elites

2025-04-19

AI Summary

The relationship between prompt structure and task performance remains poorly understood in large language model (LLM) prompt engineering. Method: This paper proposes an evolutionary prompt-space exploration method integrating context-free grammars (CFGs) with MAP-Elites. It introduces MAP-Elites—the first application of behavioral phenotyping to prompt optimization—organizing prompts into a behaviorally structured archive based on interpretable dimensions such as number of examples and reasoning depth. CFGs enforce syntactic constraints during structured prompt generation and mutation. Results: Evaluated on seven BigBench Lite tasks across multiple LLMs, the method significantly improves task-specific adaptation and cross-task prompt generalization. It enhances structural diversity while maintaining high output quality, yielding an interpretable, reusable, and systematic paradigm for prompt design.

Technology Category

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

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Prompt engineering is essential for optimizing large language models (LLMs), yet the link between prompt structures and task performance remains underexplored. This work introduces an evolutionary approach that combines context-free grammar (CFG) with the MAP-Elites algorithm to systematically explore the prompt space. Our method prioritizes quality and diversity, generating high-performing and structurally varied prompts while analyzing their alignment with diverse tasks by varying traits such as the number of examples (shots) and reasoning depth. By systematically mapping the phenotypic space, we reveal how structural variations influence LLM performance, offering actionable insights for task-specific and adaptable prompt design. Evaluated on seven BigBench Lite tasks across multiple LLMs, our results underscore the critical interplay of quality and diversity, advancing the effectiveness and versatility of LLMs.

Problem

• Exploring link between prompt structures and LLM task performance
• Systematically mapping prompt space using CFG and MAP-Elites algorithm
• Analyzing impact of structural variations on LLM effectiveness

Innovation

• Evolutionary approach with CFG and MAP-Elites
• Systematically explores diverse prompt structures
• Analyzes prompt-task alignment via key traits