Answer:
The paper introduces a new prompting method called Progressive-Hint Prompting (PHP) to enhance reasoning in Large Language Models (LLMs). PHP utilizes previously generated answers as hints to guide the model towards the correct answer in a step-by-step process. By combining PHP with existing methods like Chain-of-Thought (CoT) and self-consistency, the accuracy in reasoning tasks is significantly improved, achieving state-of-the-art performance on various benchmarks with different LLMs. The quality of hints plays a crucial role in model performance, with more powerful prompts leading to better results. Additionally, PHP reduces the number of interactions required and gradually guides the model to the correct answer, resulting in improved performance.