This survey reviews and classifies prompting techniques for Large Language Models (LLMs) into seven categories, providing a standardized framework to simplify prompt engineering. The paper aims to enhance the practical application of LLMs by offering a structured approach to understanding and utilizing these techniques across various domains.
The paper surveys prompt engineering, a technique for enhancing language and vision-language models without modifying core parameters. It categorizes recent advancements, detailing methodologies, applications, and key datasets. The paper also highlights strengths, limitations, and provides a taxonomy, offering clarity and identifying future research opportunities in this rapidly evolving field.
The “Tree of Thoughts” (ToT) framework improves language model inference by enabling exploration of multiple reasoning paths, allowing for strategic decision-making and backtracking. ToT significantly boosts performance on tasks like Game of 24, where it achieved a 74% success rate compared to GPT-4’s 4% using traditional methods, enhancing problem-solving capabilities.
Tree of Thoughts: Deliberate Problem Solving with Large Language Models Read More »
The paper demonstrates that generating a “chain of thought”—a series of intermediate reasoning steps—significantly enhances large language models’ performance on complex reasoning tasks. Through chain of thought prompting, where exemplars are used in prompts, the method shows substantial improvements across arithmetic, commonsense, and symbolic reasoning tasks. Notably, a 540B-parameter model, prompted with eight such exemplars,
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models Read More »