Core Concepts
A novel auto-prompt graphical paradigm that combines stimulating and framework prompts to enhance the problem-solving capabilities of large language models across multiple domains.
Abstract
The paper proposes an Auto-Prompt Graphical Paradigm (APGP) that integrates two types of prompts - stimulating prompts and framework prompts - to improve the problem-solving abilities of large language models (LLMs).
The key highlights are:
Categorization of traditional prompts into stimulating prompts and framework prompts, and the introduction of a new prompt type that combines the advantages of both.
Design of the APGP, which automates the prompt design process and incorporates emotional stimuli to guide LLMs through problem abstraction, solution generation, optimization, and self-verification.
Development of a framework to instantiate the APGP, demonstrating its effectiveness on the Ruozhiba and BIG-Bench Hard datasets.
Ablation studies confirming the importance of the stimulating prompts in the framework and the potential for further optimization.
The framework aims to leverage the universality of stimulating prompts and the task-specific features of framework prompts to better exploit the latent capabilities of LLMs in solving complex problems across multiple domains.