Wang et al. (2023), "Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models," target error categories that plague Kojima et al.'s zero-shot "Let's think step by step" prompting, such as missing reasoning steps and calculation errors. What does Plan-and-Solve prompting add to address this, without using any worked examples?
- It instructs the model, in a single zero-shot prompt, to first devise a plan that divides the overall task into smaller subtasks, and then to carry out that plan step by step before giving the final answer
- It requires collecting several hundred worked examples with expert-annotated plans and fine-tuning the model on them before it can perform any reasoning
- It replaces natural-language reasoning entirely with a formal, executable program that a separate code interpreter runs to produce the final answer
- It asks the model to skip straight to the final numeric answer without showing any intermediate reasoning, in order to avoid calculation errors introduced by long chains of text
Why A? And why not the others?
Correct answer: A. It instructs the model, in a single zero-shot prompt, to first devise a plan that divides the overall task into smaller subtasks, and then to carry out that plan step by step before giving the final answer
Plan-and-Solve prompting keeps Zero-shot-CoT's no-worked-examples format but adds an explicit planning instruction: the model is prompted to first devise a plan that breaks the overall problem into smaller subtasks, and then to execute that plan step by step before stating its final answer, with an extended "PS+" variant adding further instructions to extract relevant variables and pay attention to calculation accuracy. This targets exactly the missing-step and semantic-misunderstanding errors the paper identifies in plain Zero-shot-CoT. The second option is wrong because Plan-and-Solve remains a purely zero-shot, inference-time prompting method with no fine-tuning and no annotated example collection. The third option is wrong because that describes Program-of-Thought prompting, a different technique the paper compares against; Plan-and-Solve itself keeps natural-language reasoning and simply adds an explicit planning step. The fourth option is wrong because the technique asks for more structured intermediate reasoning, not less, and the paper's goal is reducing errors within that reasoning, not removing it.
Source: Wang, Xu, Lan, Hu, Lan, Lee, Lim, "Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models" (ACL 2023), arXiv:2305.04091