Lecture

Rome Wasn't Built in a Day, Understanding Chain of Thought Prompting

Chain of Thought prompting is a method that helps AI solve complex problems step-by-step while guiding it to explain its reasoning process. It is commonly abbreviated as CoT.

CoT not only involves asking AI to find an answer but also requires a step-by-step explanation of how that answer is reached. Often, phrases like step-by-step or in order are included at the end of a prompt to encourage sequential reasoning.

CoT prompting was concretized in a 2022 Google research paper, Chain of Thought Prompting Elicits Reasoning in Large Language Models. According to the paper, CoT prompts improved the accuracy of responses from the GPT-3 model from 38.0% to 54.5%.

This approach is particularly effective in improving outcomes for problems requiring mathematical reasoning by analyzing problem scenarios step-by-step.


How Can I Use It?

CoT prompting includes multiple scenarios within a single prompt similar to Few-Shot Prompting, but it guides each scenario to be reasoned out and resolved step-by-step.

CoT Prompt Example
You received $500 as a business travel budget. You spent $200 on hotel accommodation, and $100 on transportation. From the remaining amount, you plan to use $20 per meal. How many meals can you have at most? Please calculate it step-by-step.
AI Response Example
Starting Point: You received $500 as a business travel budget. Total Budget: $500 First Expense Review: You spent $200 on hotel accommodation. Remaining Budget Calculation: $500 - $200 = $300 Second Expense Review: You spent $100 on transportation. Remaining Budget Calculation: $300 - $100 = $200 Remaining Amount Check: After spending on hotels and transportation, you have $200 left. Current Remaining Budget: $200 Unit Cost Confirmation: You plan to use $20 per meal. Cost per Meal: $20 Maximum Number of Meals Calculation: Calculate the maximum number of meals possible with the remaining money. Calculation Method: Remaining Amount ÷ Cost per Meal = $200 ÷ $20 Final Calculation and Result: Calculation Result: $200 ÷ $20 = 10 meals Maximum Number of Meals: 10 meals Answer: You can have a maximum of 10 meals.

CoT excels in solving mathematical problem scenarios, as shown above, but it can also be applied to general situations requiring step-by-step analysis, as seen below.

CoT Prompt Example 2
You are preparing for a marketing plan presentation for summer air conditioner sales. Provide the necessary content for each step of the presentation preparation. 1. Research 2. Must-Include Content in Presentation Slides 3. Points to Note During the Presentation 4. Presentation Script
AI Response Example 2
1. Research - Market Analysis: current market size for air conditioners, growth rate, major trends - Competitor Analysis: main competitors' product features, price range, marketing strategies - Consumer Analysis: consumer preferences, purchase factors, key target groups (age, region, income level) - Product Analysis: key features, advantages, and technical superiority of the company's air conditioners - Past Data Analysis: previous summer air conditioner sales data and effective marketing channels 2. Must-Include Content in Presentation Slides - Market and Consumer Analysis: Market status, target consumer analysis - Competitor Comparison: Comparison with main competitors' products - Marketing Goals: Specific sales goals, market share goals - Marketing Strategy: - Product Strategy: Product USP (Unique Selling Proposition) - ...

Thus, CoT prompting uses crucial keywords like step-by-step, in order to direct AI toward sequential reasoning and can be effectively applied in solving mathematical problems and general situations requiring step-by-step analysis.


Practice

Send a prompt example and compare the AI's response.

Mission
0 / 1

Which of the following best describes a characteristic of Chain-of-Thought (CoT) prompting?

A method that asks the AI to perform a specific task without examples

A method that guides the AI to solve complex problems step-by-step

A method that provides the AI with numerous examples

A method that restricts the length of the AI's response

Lecture

AI Tutor

Publish

Design

Upload

Notes

Favorites

Help