Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models

By Lei Wang, Wanyu Xu, Yihuai Lan, Zhiqiang Hu, Yunshi Lan, Roy Ka-Wei Lee, Ee-Peng Lim. Annual Meeting of the Association for Computational Linguistics, 2023

783 citations recorded by the source corpus.

Large language models (LLMs) have recently been shown to deliver impressive performance in various NLP tasks. To tackle multi-step reasoning tasks, Few-shot chain-of-thought (CoT) prompting includes a few manually crafted step-by-step reasoning demonstrations which enable LLMs to explicitly generate reasoning steps and improve their reasoning task accuracy. To eliminate the manual efforts, Zero-shot-CoT concatenates the target problem statement with “Let’s think step by step” as an input prompt to LLMs. Despite the success of Zero-shot-CoT, it still suffers from three pitfalls: calculation errors, missing-step errors, and semantic misunderstanding errors. To address the missing-step errors, we propose Plan-and-Solve (PS) Prompting. It consists of two components: first, devising a plan to divide the entire task into smaller subtasks, and then carrying out the subtasks according to the plan. To address the calculation errors and improve the quality of generated reasoning steps, we extend PS prompting with more detailed instructions and derive PS+ prompting. We evaluate our proposed prompting strategy on ten datasets across three reasoning problems. The experimental results over GPT-3 show that our proposed zero-shot prompting consistently outperforms Zero-shot-CoT across all datasets by a large margin, is comparable to or exceeds Zero-shot-Program-of-Thought Prompting, and has comparable performance with 8-shot CoT prompting on the math reasoning problem. The code can be found at https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/62176de125738e3b95850d1227bac81fd646b78e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

When large language models meet personalization: perspectives of challenges and opportunities (2023, positive): “Subsequent studies like ToT [228], Plan-and-Solve [229] and ReAct [230] have proposed more advanced techniques for prompt design to assist in guiding LLM to engage in deeper thinking and tool planning.”
Understanding the planning of LLM agents: A survey (2024, background): “Plan-and-Solve [ Wang et al. , 2023b ] improves upon the Zero-shot Chain-of-Thought [ Kojima et al. , 2022 ] by transforming the original “ Let’s think step-by-step ” into a two-step prompt instruction: “ Let’s first devise a plan ” and “ Let’s carry out the plan ”.”
To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning (2024, methodology): “Many symbolic and semi-symbolic tasks be broken down into two stages (Ye et al., 2023; Pan et al., 2023; Jiang et al., 2024): planning, either via a formal or informal specification via prompting (Sun et al., 2024; Wang et al., 2023b), and execution, using the same LM or external solvers.”
Reasoning on Graphs: Faithful and Interpretable Large Language Model Reasoning (2023, background): “To further unleash LLMs’ reasoning ability, the plan-and-solve paradigm (Wang et al., 2023c) has been proposed, in which LLMs are prompted to generate a plan and execute each reasoning step.”
Tree of Thoughts: Deliberate Problem Solving with Large Language Models (2023, background): “As they are trained on vast amount of world knowledge and human examples, LMs are known to have already absorbed rich commonsense that makes it possible to propose reasonable plans conditioned on problem setting and environmental states [10, 39, 34, 11, 32, 38, 37].”
LLM-Assisted Code Cleaning For Training Accurate Code Generators (2023, methodology): “This corresponds to various LLM instruction or fine-tuning approaches (Xu et al., 2023; Li et al., 2023b) providing a strong baseline for data cleaning.”
MAmmoTH: Building Math Generalist Models through Hybrid Instruction Tuning (2023, methodology): “More recently, (Chen et al., 2023; Wang et al., 2023e) have proposed to tackle more challenging college-level science and math problems.”
Measuring Faithfulness in Chain-of-Thought Reasoning (2023, background): “Approaches include generating subquestions (Dua et al., 2022; Zhou et al., 2023), producing a tree of thoughts (Yao et al., 2023a), devising and executing a plan for answering the question (Wang et al., 2023), and having language models debate to reach an answer (Du et al., 2023), among others.”
Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs (2024, positive): “This has been expanded with methods like debating, self-correction, self-critique, and plan-and-solve (Liang et al., 2024; Du et al., 2024; Xiong et al., 2023; Kumar et al., 2024; Kamoi et al., 2024; Ke et al., 2023; Lin et al., 2024; Yu et al., 2024; Wang et al., 2023a).”
Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs (2023, background): “Alternative strategies such as plan-and-solve (Wang et al., 2023) can also be used.”
Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks (2022, background): “…problems with PoT. self-critic (Gou et al., 2023) and self-eval (Xie et al., 2021) both adopt self-evaluation to enhance the robustness of the generated program. plan-and-solve (Wang et al., 2023a) instead adopt more detailed planning instruction to help LLMs create a high-level reasoning plan.”
WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct (2023, positive): “Figure 4 illustrates that the evolution process does not yield higher similarity scores.”
Large Language Model (LLM) for Telecommunications: A Comprehensive Survey on Principles, Key Techniques, and Opportunities (2024, methodology): “In plan generation, LLM models can utilize text-based approaches to produce natural language sequences or code-based methods for generating executable programs, enhancing the verifiability and precision of the planned actions [77].”
ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs (2023, background): “In summary, our primary contributions are: (Kojima et al., 2022; Wei et al., 2022; Wang et al., 2023a) and Tree-of-Thought prompting (Yao et al., 2023a), self-consistency (Wang et al., 2023b), meta-reasoning over multiple paths (Yoran et al., 2023), use of scratchpads (Nye et al., 2021), training…”