Diverse Prompts: Illuminating the Prompt Space of Large Language Models with MAP-Elites
This work addresses the problem of optimizing prompt design for LLM users by providing systematic insights into prompt structures, though it is incremental as it builds on existing evolutionary and grammar-based methods.
The paper tackled the underexplored link between prompt structures and task performance in large language models by introducing an evolutionary approach using context-free grammar and MAP-Elites to generate diverse, high-performing prompts, revealing how structural variations influence performance across seven BigBench Lite tasks.
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.