Chain of Draft: Thinking Faster by Writing Less

2025-02-25

AI Summary

Large language models (LLMs) incur high computational overhead and latency in complex reasoning due to reliance on lengthy chain-of-thought (CoT) prompting. Method: This paper introduces Chain of Drafting (CoD), a novel reasoning paradigm inspired by human cognitive refinement of drafts. CoD guides LLMs to generate minimal yet information-complete intermediate reasoning steps—retaining only essential logical elements—via prompt engineering, without fine-tuning or additional parameters, and with broad compatibility across open- and closed-source LLMs. Contribution/Results: On diverse reasoning benchmarks, CoD matches or exceeds CoT accuracy while reducing inference token consumption to 7.6%–32% of CoT’s usage, substantially lowering computational cost and response latency. To our knowledge, this is the first work to formalize human draft-based reasoning cognition into a lightweight, deployable LLM inference paradigm.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs) have demonstrated remarkable performance in solving complex reasoning tasks through mechanisms like Chain-of-Thought (CoT) prompting, which emphasizes verbose, step-by-step reasoning. However, humans typically employ a more efficient strategy: drafting concise intermediate thoughts that capture only essential information. In this work, we propose Chain of Draft (CoD), a novel paradigm inspired by human cognitive processes, where LLMs generate minimalistic yet informative intermediate reasoning outputs while solving tasks. By reducing verbosity and focusing on critical insights, CoD matches or surpasses CoT in accuracy while using as little as only 7.6% of the tokens, significantly reducing cost and latency across various reasoning tasks.

Problem

• Enhance reasoning efficiency
• Reduce token usage
• Minimize computational cost

Innovation

• Minimizes token usage
• Enhances reasoning efficiency
• Reduces computational costs