Foreword: It's the power of human thinking that gives wings to artificial intelligence. To improve your productivity, hurry up and see, an article that teaches you how to ask chatgpt the right questions.

"Finally using ChatGPT!"

"Eh? Why does it always fail to answer the questions I ask?"

"It doesn't feel that good?"

Newbies just starting out with ChatGPT often encounter situations where the answer is not the answer, and even very simple math questions and search questions can't get an accurate answer. However, when others are using it, chatgpt always gives a very accurate answer, to writing beautiful academic articles and dealing with difficult interviews. In fact, mastering some questioning skills, we can get quality answers exactly like others, and really let chatgpt assist our own work, improve our own efficiency and find inspiration for our daily life.

First of all, we need to figure out why ChatGPT is "wrong".

We should know that chatgpt is not a human who learns all the knowledge, it is a machine learning based technology, through a lot of training and testing, to be able to answer more users' questions.

If we ask chatgpt about it ，he will say four reasons for errors:

1, Lack of information : we can understand that, on the one hand, for the Internet information is not accurate, on the other hand, also because the current chatgpt training using data is not real-time, for the data before 2022, so some real-time problems, the model does not understand.

2. Misinterpretation of context: we can understand that the questioner did not ask the question accurately.

3. Limitations of language models: the model does not fully understand some of the terminology that needs to be understood.

4, iTechnical issues: the model still needs to be optimized and is also evolving.

Understanding these four points, then we can optimize the way we ask questions, but need to pay attention to.

Although in this article we are going to introduce several better ways of asking questions to improve the accuracy of the replies received, there is no way that has been proven in the data to allow ChatGPT to make 100% accurate answers.

We will always need to make human judgments about ChatGPT's responses.

We will always need to make human judgments about ChatGPT's responses.

We will always need to make human judgments about ChatGPTs responses..

A more optimal way to ask questions

I split the simple question into "set requirements"-"limited scope"-"split questions"-"optimized responses" four steps to optimize our answers, which is a way of the most basic training.

Simply put, first let ChatGPT understand which aspect of the question you want to ask? From what angle of analysis? What form of response do you want to get? After determining everything, after getting the response, keep optimizing the response to get the answer you want. Next we distance instructions.

1、Set up requirements

The basic settings include language settings, for example, enter "Respond in English" to get an English reply.

In addition, we can set some custom conditions in advance, such as setting abbreviations, next, if I send the content of {League}, I mean {League of Legends the game}.

We can also set some constraints, for example, in the next translation, don't translate the word "prompt". etc.

Or, we can set the meaning of a word in advance, for example, next, "dawn" represents a person's name, not the concept of time.

2. Limit the scope

After the basic setting, we can give the AI an identity and background, and specify the range of answers, style, and more needs you need to reply to.

For example, "I would like you to act as an IT specialist. I will want you to provide all the information about my technical problem, and your role is to solve my problem. You should use your knowledge of computer science, network infrastructure, and IT security to solve my problem. In your answer, use smart, simple language that can be understood by people of all levels. And explain step-by-step your solution and the key points of use. Try to avoid too many technical details, but use them when necessary. I want you to answer with a solution, not any parsing."

Let's break down the conversation above

a. We give the AI an identity, "I want you to act as an **IT expert**. I will want you to provide all the information about my technical problem, and your role is to solve my problem".

b. limited the scope of the answer to the question, "You should use your knowledge of computer science, network infrastructure, and IT security to solve my problem." .

c. specifies the way to answer the question, "In your answer, use smart, simple language that can be understood by people at all levels. and explain step-by-step your solution and the key points used. Try to avoid too many technical details, but use them when necessary. I want you to answer in solution, not any parsing."

In this way, you can make the AI understand more clearly the question you want to ask, of course, this is for complex questions, if it is a simple question, just focus on one or two of the parts to get a precise answer.

3, Split Questions (stage questions)

Because in chatgpt is the ability to combine context, as long as in a single conversation, we can train ChatGPT to understand our ideas by asking questions in succession.

So, for some more complex problems, we use staged questioning, for example, when we want ChatGPT to assist us in making a PowerPoint, writing an essay, designing various programs, we can split the production of content into,

"determining the scope of the topic"," making an outline", “generating copy for different chapters”, “generating summarized content”, etc. (splitting the big problem into smaller, more defined problems can effectively (improve the accuracy of AI).

For example, we want to make a PowerPoint of 2019 China game industry analysis, I can split the problem into these steps (just a simple example, it does not represent the optimal answer.)

a. I want to make a PowerPoint of 2019 China game industry analysis, from which angles can we analyze?

b. For the analysis angles just proposed, I want to organize into a 30-page ppt, can you help me plan the ppt outline of 2019 China game industry analysis including these 7 directions, mark clearly what is said in each page of the ppt.

c. Share with me 10 references that may be used in this PowerPoint, including their web links.

d. According to the outline you wrote, explain "Game Market Overview" in detail with 200-300 words of copy.

...... (according to the outline, let ai write out the content of each page of the PowerPoint in detail)

When we have done the above three steps, the last step is to optimize the content.

4, optimize the response

Filtering and optimizing the content is an essential step, and a very critical step in training a model. For example, we can let chatgpt provide more and broader content, we can also let ChatGPT simplify the content, or let it use different display methods.

Here are some common optimization languages, "optimize the above to less than 50 words" "give me another solution" "refactor the code to show the changed content in vimdiff format and explain the changed content."

Summary

By asking questions through the above four steps, we can effectively improve the accuracy of our questions. If we really don't know how to ask questions to ChatGPT and reap more accurate results, we can also ask questions to ChatGPT directly: "I want to ask a question about car repair, what kind of way should I ask questions to ChatGPT to What is the most accurate way to ask ChatGPT about a car repair question?"

If you are too lazy to think about it, then in the next article, I will share some common questioning models for some common questions for your daily use.