In the ever-changing realm of prompt engineering, efficiency isn’t just a luxury—it’s a necessity. Traditional techniques like Standard and Chain-of-Thought (CoT) have undoubtedly done wonders, but they can be like writing an essay when a tweet would do. Enter Chain-of-Draft (CoD) prompting, a fresh approach inspired by how we naturally jot down only the essentials. This post unpacks CoD, revealing its benefits with vivid examples and analogies, contrasting it with other methods, and offering a clear, engaging roadmap for when to use each technique.
What Is Chain-of-Draft Prompting?
Imagine solving a puzzle by scribbling only the key hints on a sticky note rather than writing a full paragraph for every thought. Chain-of-Draft prompting does just that—it compacts the reasoning process into crisp, bite-sized steps. According to the Chain of Draft: Thinking Faster by Writing Less paper, by limiting each reasoning step to about five words, CoD dramatically cuts down on token usage and speed bumps, without a steep hit to accuracy .
Why CoD Prompts Are a Game-Changer
• Lightning-Fast Responses:
CoD trims the fat by cutting the number of tokens, reducing latency like a well-oiled machine—perfect for real-time tasks.
• Cost-Saving Brilliance:
With fewer tokens, you’re not burning through your computational budget. Think of it as using a fuel-efficient car for your long journey.
• Smart and Lean Accuracy:
Despite its brevity, CoD holds its ground with accuracy nearly rivaling the detailed CoT approach, and sometimes even outshining it in symbolic tasks .
• No More Overthinking:
By zeroing in on the essentials, CoD avoids the pitfalls of over-complication and those pesky hallucinations that verbose models can suffer from.
Real-World Examples: Showcasing Effective CoD Prompts
Here are some clear, stand-alone examples that demonstrate how to craft effective Chain-of-Draft prompts—each with succinct reasoning steps and a final answer.
Example 1: Arithmetic Problem
Prompt:
“Using a Chain-of-Draft approach, solve the following problem. Each reasoning step should be no more than 5 words. Provide the final answer after ####.
Problem: Lisa has 30 apples, gives some to Mark, and is left with 18 apples. How many did she give away?”
Expected CoD Response:
“30 – x = 18;
x = 12.#### 12”
Explanation: The prompt instructs the model to focus on the essential equation, resulting in a clear, concise breakdown that directly leads to the answer.
Example 2: Logical Deduction
Prompt:
“Apply Chain-of-Draft reasoning to this puzzle. Limit each step to 5 words or fewer, and place the final conclusion after ####.
Problem: All roses are red. Some flowers are roses. What can be inferred about some flowers?”
Expected CoD Response:
“Roses are red;
Some flowers are roses;
Inference: Some flowers red.#### Some flowers are red.”
Explanation: This prompt requires the model to distil logical deductions into short, direct statements that reveal the final conclusion without unnecessary detail.
Example 3: Symbolic Reasoning
Prompt:
“Using a Chain-of-Draft method, determine the truth of this statement. Each reasoning step must not exceed 5 words, with the final verdict following ####.
Problem: If A implies B and B implies C, does A imply C?”
Expected CoD Response:
“A implies B;
B implies C;
Thus, A implies C.#### True”
Explanation: The prompt guides the model to break down a logical implication into its core parts, ensuring clarity and brevity, and directly presenting the conclusion.
These examples illustrate how Chain-of-Draft prompts can transform complex problems into clear, minimal reasoning processes, making it easier to follow the logic and arrive at the answer quickly.
When to Use Each Technique
• Simple Queries:
Stick with Standard prompting for no-fuss, direct answers.
• In-Depth Explanations:
CoT is your go-to when you need every detail on display.
• Efficiency and Quick Turnarounds:
CoD is perfect for real-time applications and cost-sensitive projects.
• Complex or Multi-Threaded Tasks:
Consider alternatives like Skeleton-of-Thought when tasks allow for parallel processing—but only if you’re ready for a bit more complexity.
Conclusion: Embrace the Minimalist Revolution
Chain-of-Draft prompting is a refreshing leap forward—it captures the essence of human thinking: concise, direct, and effective. By cutting the clutter, it delivers almost all the benefits of detailed reasoning with a fraction of the cost and time. If you’re keen on balancing clarity with efficiency, CoD is a technique worth mastering. Ready to supercharge your prompt engineering toolkit and join the minimalist revolution?
Let’s get those ideas flowing—what’s your next move?
ChatGPT Notes:
In this engaging collaboration, Manolo and I (ChatGPT) crafted a compelling blog post exploring Chain-of-Draft (CoD) prompting, a minimalist yet powerful approach to AI reasoning.
• Manolo provided:
• The research paper introducing CoD prompting
• Clear guidance on structuring the post for prompt engineers
• Feedback to refine tone, improve flow, and enhance readability
• Requests for stronger real-world examples and more engaging storytelling
Through multiple revisions, we refined the content to balance technical depth with accessibility. Manolo also used AI to generate the accompanying blog image.