Make good use of thought chains to improve the questioning efficiency of ChatGPT

[](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj9CSmZwvEH6MWeX_MHrancGGS-M4GcAO4-YGaT_Bk0iuXZctkSKMp7SRNvXmn XXlsOf5g9VKm99In7vf3jFyfAVEodUImvEFx_ZMZnrLoXiwGWYZjjHFydTdbICWwpItkmxfYoRhGFDPQIyPEFY5zcJgVnHC5hDOlAuoLjq2CkzfDOrtM865Ol1Onrj R6I/s1792/DALL%C2%B7E%202024-01-26%2009.28.51%20-%20A%20young%20girl%20reading%20and%20contemplating%20in%20a%20beautiful,%20 ethereal%20style.%20The%20illustration%20should%20be%20imbued%20with%20an%20aesthetic%20charm,%20featuring%20soft,%20drea.png)

This article is written by Zheng Weiquan and was originally published in “Technice Island”

The time sequence has entered 2024. First of all, I would like to say Happy New Year to everyone! I believe that during the past year, everyone has become quite familiar with generative AI tools such as ChatGPT, Bard or Claude! However, what if you are asked whether you are satisfied with the answers provided by ChatGPT? I think many people may smile bitterly, right?

Now, let me share with you a questioning technique called “Chain of Thought” (Chain of Thought, CoT). Simply put, it’s a way for robots to learn to think like humans. Just like when we were solving a math problem as children, the teacher would teach us to write down the problem-solving process step by step, and the “thinking chain” will allow the robot to imitate this reasoning process.

In other words, if you ask ChatGPT a complex mathematical problem in this way, it will start “thinking” and tell you step by step how it solves the problem, instead of jumping directly to the final answer.

Two major benefits of thinking chain

Doing so has two obvious benefits: First, it makes ChatGPT more likely to find answers when faced with truly complex problems. Secondly, it also makes it easier for us to understand what ChatGPT thinks? If the answer is wrong, we can more easily see where the problem lies?

It’s like when you discuss with a friend where to eat, he may quickly recommend a restaurant with a high CP value to you based on your preferences and budget. Next time, when you ask him related questions, he will naturally provide suggestions based on the content of the last conversation.

According to Wikipedia’s Introduction, “Thinking Chain” can also be called “Thinking Chain”. It is a technology of text prompting (Textual Prompting), which prompts large language models (Large Language Models (LLM) improve their reasoning capabilities by generating a series of intermediate steps that lead to the final answer to a multi-step problem. This technology was first proposed by researchers at Google in 2022.

According to research scientists Jason Wei and Denny of the Google Brain team [Research] by Zhou et al. (https://arxiv.org/abs/2201.11903), “Thinking Chain” plays a significant role in the process of training large language models and designing question words.

Some friends may have heard of the term “Prompting Engineering”, which is a technology that allows generative AI tools such as ChatGPT to better understand how to answer questions. It provides questions in the form of prompt words to the AI ​​tool for reference, instead of just providing parameters. In this way, AI tools can learn directly from the problem itself.

Scientists have found that adding a sentence like “Let’s think step by step” to the prompt word can improve the performance of generative AI tools such as ChatGPT on problems that require multi-step reasoning. In other words, this is the so-called “thinking chain” technology.

“Thinking Chain” allows AI tools to generate a series of reasoning steps before answering a question. Its principle is just like the process of the human brain when thinking. In this way, the ability of AI tools to solve complex problems can be improved, especially when faced with problems that require thinking about reasoning or mathematical calculations.

Trying to expand the scale of language models is certainly a technological breakthrough, but the process is not only long, but also usually expensive. Therefore, a group of experts and scholars decided to find another way to use “thinking chain” technology to effectively improve the performance of AI tools such as ChatGPT on complex reasoning problems. It is good at simulating the human thinking process, which is enough to make the answers of AI tools easier to understand.

How the thinking chain works

“Thinking chain” sounds a bit esoteric, but in fact it refers to a series of logically related thinking steps, but we connect them in series to form a complete thinking process.

Please think about it, when we face some complex problems in life, how would you dismantle and deal with them? For myself, I like to use a tool like “Mind Map” to comprehensively break down the thinking steps. For example, when you need to make a briefing or prepare a speech, we will first break the topic into related subtopics, then gradually think about and implement the details for each subtopic, and finally organize all the ideas to form a clear context. This entire process from decomposition to organization is a chain of thinking.

Overall, “Mind Map” is a very suitable tool for dismantling and reconstructing knowledge, and “Thinking Chain” is also a great thinking tool! Therefore, in the process of learning prompt word design, I also suggest that you can use the “thinking chain” method to prompt the reasoning of large language models.

This step-by-step questioning method makes it easier for the questioner to check whether the reasoning of the large language model is correct? Even if mistakes are made during the questioning process, it doesn’t matter. We can correct them in time.

In short, using the concept of “thinking chain” to design prompt words is like asking ChatGPT, Bard or Claude to do a detailed mathematical analysis question, not just a simple fill-in-the-blank question. It allows these generative AI tools to display the reasoning process and the logic of each step in detail, thereby improving their reasoning capabilities.

The origin of “Think Chain” can be traced back to the Socratic question-and-answer method in ancient Greece. This method explores the topic in depth through a series of questions and answers. In modern times, this method is also widely used in various fields such as management consultants, education, psychology, and scientific research.

Features of “Thinking Chain” include:

• Coherence: There is logical coherence between questions, and each question builds on the answer to the previous question.
• Depth: Through a series of questions and answers, Thought Chain allows for in-depth exploration of multiple aspects of an issue.
• Flexibility: Can be applied to almost any type of problem-solving process, regardless of topic or domain.

The importance of “thinking chain” is self-evident, mainly because it provides a structured way of thinking that can help people understand the problem more comprehensively and deeply. This approach is particularly suitable for solving complex problems because it encourages exploring different angles of the problem and uncovering new insights and solutions.

Application cases for coffee shop entrepreneurship

Finally, let me give you a common case in the workplace to help you understand how the “Thinking Chain” is applied.

Suppose Wang Jiahua is a young man who wants to start a business, and he wants to use ChatGPT to get market analysis and advice on opening a new coffee shop.

He can use the concept of “thought chain” to ask questions to ChatGPT:

1. Start asking: What is the current market situation of coffee shops in downtown Taipei? ChatGPT may provide information about the number, type, and general footfall of coffee shops in downtown Taipei.
2. In-depth questions: Who are the main customers of these coffee shops? Use this to understand the type of customers that different coffee shops attract, such as office workers, students or tourists.
3. Further exploration: How have the preferences of coffee consumers in Taipei changed in recent years? This helps Wang Jiahua understand market trends, such as customer preferences for special coffee beans.
4. Market opportunities: Are there currently unmet customer needs in the market? Explore potential market opportunities, such as specialty beverages or unique coffee experiences.
5. Competitive analysis: What is the competitive landscape of coffee shops in downtown Taipei? Understand your competitors’ strengths and weaknesses, and how saturated the market is.
6. Potential Opportunities: If opening a new coffee shop, what are the key factors to consider? Evaluate the impact of location, store design, menu selection and other factors on store opening.

Through this question-and-answer process, Wang Jiahua can not only obtain market information about coffee shops, but also gain an in-depth understanding of the trends in related industries and explore potential business opportunities.

Now, I believe you already understand: “Think Chain” is a problem-solving method that allows you to explore a topic in depth through continuous questioning. Each question builds on the answer to the previous question, allowing you to drill down to the heart of a topic or uncover deeper insights.

★ Want to embrace the AI era? Welcome to sign up for “AI Copywriting Course”!

★ To comprehensively improve your writing skills, welcome to sign up for the “Vista Writing Companion Program”.

---