Pranab Sahoo, Ayush Kumar Singh, Sriparna Saha, Vinija Jain, S. Mondal, Aman Chadha
A comprehensive survey that systematically categorizes prompt engineering techniques by application area and analyzes the strengths and limitations of each method.
Prompt engineering is an important technique for extending the capabilities of LLMs, but various methods have been proposed sporadically, making systematic understanding and comparison difficult. Researchers struggle to determine which prompting technique is suitable for which task and what the pros and cons of each technique are, hindering efficient research progress.
This paper categorizes recent prompt engineering research by application area (e.g., question answering, commonsense reasoning, text generation) and summarizes the prompting methodology, applied models, and datasets used for each approach. It also analyzes the strengths and limitations of each technique and provides a taxonomy diagram and summary table to facilitate systematic understanding.
This paper systematically organizes the overall research landscape of prompt engineering, enabling researchers to easily grasp and compare the characteristics of each technique. Additionally, by identifying current limitations and open challenges, it suggests future research directions, contributing to the advancement of this rapidly developing field.