Reading Tips:
1. The content of the article is hardcore, you need to concentrate, it is not recommended to read in pieces, please reserve 30 minutes of the whole time.
2. This article is a complete set of knowledge system, the article will not come up and throw you a solution directly, but will start from the "what" and "why", tell you the cause and effect, and finally tell you how to do it.
Here is the text:
The birth of ChatGPT has made many people anxious, but you don't need to worry about it at all, because more powerful than AI is always the human who harnesses AI for its own use.
And GPT is far less mysterious than the hype of major merchants, and its application logic is very simple, you don't need to spend money to enroll in various classes for this.
Today I will use an article to take you to master the use of GPT, this article has no nonsense, the whole process is dry, all are practical, pure white can also understand.
As long as you can read this article carefully, I guarantee that your ability to apply GPT will surpass 90% of people!!!
Image generated by AI
So, before we officially begin, it is necessary for me to explain here:
About the basic introduction of GPT, and how to register, these search engines search a lot of things, we don't waste tongues here.
PS: In order to take care of some students without special environment, the article will also provide some mirror websites that are directly accessed in China, please pay attention to find Easter eggs~
Therefore, we will directly focus on the most critical part, that is: the two major pain points after you get GPT....
1
Why can't AI work for you?
The first pain point after many students get GPT is: it is not used well.
That said, you often feel that ChatGPT's answer is empty and doesn't have much reference value...
The second pain point is: nowhere to use it.
That is to say: after having GPT, I find that I don't seem to have anything to ask, and I don't know what I can do with GPT.
Therefore, under the influence of these two pain points, when you toss the account and the fresh period of idle chat passes, GPT will lie flat and eat ashes, and your life and work will remain unchanged.
However, my results with AI may not be the same as yours~
It can be said that everything in my current workflow has basically been taken over or assisted by AI, which has more than tripled my overall efficiency.
It has really become my personal assistant, permeating every aspect of my work, study, and life....
Including the article you are seeing now, in addition to the writing is done by Chang Chang himself, other links such as typesetting, addition, editing, error correction, etc., are all done through AI....
In fact, with the same tool, I can use it to create real value, but you fall into the core reason of the embarrassing situation of "answer well, I don't know what to do".
It's because you're missing these two points~
1. You got the dragon slaying knife, but don't know how to swing it.
That is: not mastering the use of AI.
2. You can swing it, but you don't know where to swing and where to slay the dragon.
That is: it is impossible to associate AI with its own application scenarios, and finally the dragon slaying knife can only be used as a fire stick.
It can be said that almost all students who cannot use AI to create real productivity have died at these two levels.
Therefore, the logic of our next explanation is around: the method of using AI (helping you swing the dragon slaying knife).
and the two logical lines of correlation AI application scenarios (let you know where to slay the dragon) to help you thoroughly master the application of GPT~
First of all, let's first introduce the first line of using AI, that is: take you to master the method of using AI to help you [swing the AI dragon slaying knife]!!
2
So how to use AI?
In fact, the core mental method of the whole gas AI can be summarized in one sentence:
In other words, the quality of the answers generated by GPT depends entirely on how you "ask it" and how you "guide it".
If you can ask well and guide well, then it will help you generate answers that surprise you, otherwise it is worthless and false.
And the "way to ask it" here refers to: the language of communication with AI, and the "way to guide it" is the method of tuning AI.
It can be said that as long as you understand these two points, then you have mastered the ability to wield the dragon slaying knife of AI.
So, what does this phrase mean?
In fact, it is easy to understand that because the current technology of human technology for natural language (human words) processing technology is not perfect, so the quality of AI-generated content is very dependent on prompts
PS: You can understand the "prompt word" as: let the AI accurately get the words you intent, or the language way to communicate with the AI.
That is to say: if the prompt words you give to AI are not of good quality or are not in place, then what AI gives you is often the kind of text garbage that is "like a pile of correct nonsense".
These things are not fundamentally different from the patchwork words you search engines and are not enlightening to you.
Therefore, the first step to get high-quality answers from AI is to first learn the language of communication with AI, that is, learn to write prompts.
Regarding the importance of writing prompts for AI, Sam Altman, CEO of OpenAI and the father of ChatGPT, specifically tweeted that writing prompts for AI is a very high-leverage skill.
Then, in order to make you more directly and clearly aware of the importance of being able to write prompts, we may wish to add another example that can be directly perceived by you~
Let you see how far the quality of the results returned to you by AI will be if you have the same demand for AI, can write prompt words, and will not write prompt words...
PS: The picture can be enlarged after clicking
Well, through the above introduction, you realize the importance of being able to write prompts, so how do we write high-quality prompts?
On this point, I will summarize a general and tried and tested prompt word template for you here~
That is to say, a good prompt word is composed of four parts: standing role + describing the problem + setting the goal + making up the requirements!!
1: [Standing role] refers to: guiding AI into specific scenarios and giving AI the status of an expert.
2: [Description] refers to telling the AI your confusion, your problem, and the background information needed to supplement the problem to the AI.
3: [Targeting] refers to: telling the AI what you need and what you want it to do for you.
4: [Supplementary request] refers to telling the AI what its answer needs to pay attention to, or what form you want it to reply to you.
Of course, this set of templates will feel more complicated to look at, but it is extremely simple to operate~
Here you may also take the previous [Travel Guide] case as an example, that brings this formula into a real application scenario, it is such an effect...
Of course, after reading this example, you may say that the parts of the template "describe the problem, set the goal, and supplement the requirements" are well understood, which is the language pattern of our usual questions.
However, why is it necessary to add a [Standing Role] action here, isn't this action a bit redundant, and it is not more simply and directly to ask questions?
Well, what I want to tell you here: this action is not only redundant, but also very important!!
For the explanation of this point, we still take the above [tourism] as an example~
Look at the same question, with and without role, how different is the answer GPT returns to you...
PS: The picture can be enlarged after clicking
Through this comparison chart, you can clearly see it~
In the case of exactly the same way and content of the inquiry, the latter only adds the six words "If you are a tour guide".
The answer returned to us by GPT is completely two realms compared to the former!!
That is to say: with the addition of questions from the role of experts, the results returned by the AI will be more specific and practical, and the tone of the answer will be more humane.
In fact, the reason for this difference is also very understandable, we might as well take Jin Yong's martial arts novel to explain~
In fact, the knowledge mastered by AI is somewhat similar to Wang Yuyan of "Dragon Babu", although she has written down the knowledge of martial arts in the world.
However, her knowledge of martial arts is broad, broad, and interferes with each other, and although she can give advice, she cannot give targeted advice.
However, when we add the role of an expert to the AI, it is no longer Wang Yuyan, who can only read dead books, but really incarnated as a practical expert in the field to give us the answer.
That is, the expert role can help the AI specify the scene, clarify the scope of the problem, and supplement the background information needed to supplement the problem.
Because an expert in a field itself represents the knowledge system of a certain field and the highest industry standards, this action is equivalent to superimposing a Buff on Wang Yuyan.
Then, on the basis of Wang Yuyan, who has mastered the knowledge of martial arts in the world, and superimposed on it an identity buff that represents the highest achievement in the field, then its answer, of course, will be very different from the answer of the nerd who only reads dead before.
So, if your question requires a certain depth and you want a more professional answer than a search engine-like answer, then remember:
The first step in asking the AI a question is to overlay it with an expert's buff, and after completing this action, tell it what you want it to do and what to ask for it.
Okay, now that you know what it means to overlay an expert buff for GPT, how do we overlay this buff on it?
In fact, the idea is very simple, after my test, you can use the following prompt words ~
You are now [xx]
Please play [XX]
If you are [XX]
Please take [XX] perspective/identity/tone.....
According to my practical experience, these prompt words can be accepted and understood by AI, and you only need to choose an expression style that best suits your language habits.
Well, when you master the method of writing prompt words, and then compare with your past way of asking, the quality of the answer given by AI will present the kind of contrast we had before~
Moreover, our idea of writing prompts is universal, and you can use this method to design prompts for almost all scenes~
◎Example 1: Use the "Prompt Word Template" to design the course syllabus
Prompt words:
AI-generated effects:
◎Example 2: Use the "prompt word template" to achieve the effect of mock interview.
Prompt words:
AI-generated effects:
◎Example 3: Use the "prompt word template" to assist the work.
AI-generated effects:
In addition, this set of ideas can also be transferred and used across scenes, such as using prompt word templates to achieve AI painting~
Prompt words:
AI-generated effects:
In short, the template is universal, the examples are endless, as long as you follow the prompt words written by this set of SOP templates, then the general GPT will not give you too bad answers.
Of course, the above is a set of standard SOP templates, if your question requirements are very simple, you don't need to follow all this set, for some simple questions, just ask directly.
Therefore, remember: the method should not be used to death, but flexible according to your actual scene needs.
Well, through the above, you will master the method of writing prompt words~
However, although it is said that through this set of ideas, high-quality prompt words can be designed to obtain high-quality answers from AI.
But we also know that current AI has not evolved to the extent of going against the sky, so for some slightly more complex questions, AI's answer is often not so in place.
So, if we want to get a better, deeper, and more valuable answer, then we need to tweak it~
Two: Tuning methods
In fact, this "method and principle of tuning" is very easy to understand~
GPT is great because Chain of Thought technology gives it the ability to have multiple rounds of dialogue and understand & combine context.
That is to say: with the blessing of this technology, the AI will remember the content of our previous conversation, and on the basis of the previous content, to answer our later content in a targeted manner, to achieve a dialogue effect similar to the communication between real people.
Therefore, based on this mechanism of AI, we can train AI by constantly "feeding data" and "casting instructions".
By constantly guiding AI, it helps us get more specific, deeper, more valuable answers, or other effects.
Well, here you know the mechanism of tuning AI, so how to train AI to make it achieve the effect you want?
Then, to achieve this effect, we need to use these two instructions.
The first instruction, I call it: continue instruction
In fact, the essence of this so-called "continue instruction" is to help you break through the output limitations of AI vendors and let the AI's answers give full play to their existence.
So for the explanation of this, we also know that the training cost of AI large models is very high.
May be based on the cost of computing power, including OpenAI, major AI vendors, will try to control the length generated by AI, and as far as possible by summarizing the text content, so that the content becomes concise.
Take ChatGPT as an example, its single maximum output will not exceed 2048 characters, as long as it exceeds this character value, the AI's answer will be forcibly truncated.
Therefore, under the two limitations of the manufacturer's space limit and the length summary, the one-time content given to us by AI will often make us feel that the content is not enough or the depth is not enough.
Then, continuing the instruction at this time can help us break through these two limitations.
That is: let the answer longer than 2048 characters continue to be answered, or let the first answer be insufficient, continue to expand in detail.
For example, take our [travel] above as an example~
When the AI reaches this position, it surpasses the characters it specifies and stops, so we can continue to let it finish the unfinished words that were not said before.
In the same way, even if it finishes speaking, we can continue to expand on the place where the answer is not deep enough and specific enough....
Of course, the examples in the text are only the most basic usage of [Continue Instruction], in addition to this most basic usage, it has further questioning usage....
For example, take the [curriculum design] mentioned above as an example~
We can use further continuation instructions to follow it up~
PS: The picture can be enlarged after clicking
Moreover, in the process of continuing to ask, we can also use the prompt words of [Supplementary Requirements] in front of us, such as:
Please explain with examples that children can understand,
Please provide no less than 5 examples,
Please select examples from the XX field,
Please answer in a lively colloquial way,
Please expand...
Please summarize...
PS: The picture can be enlarged after clicking
In theory, you can always follow the routine of "continue" and continue to ask and dig deep into it~
For example, or the [lecture example] example above, that is, you only need to follow the outline framework provided by GPT at the beginning.
Then, continue to question the content in this outline, keep matryoshka dolls, and finally fill the results of each point of inquiry into the initial big frame, so that you can get an article, which basically belongs to the courseware content generated by GPT....
The last thing you have to do is just replace the GPT-generated language with your own language style, do the logical splicing, and the final polish, and then you can lecture, well, that's it.
Of course, when using the continuation directive and extending usage, there are two more points to remind here:
Item 1: Pay attention to the ambiguity of instructions
That is to say, if your question is too long or there are too many levels of matryoshka dolls, then your continuation instructions may make the AI ambiguous and present a situation where the answer is not asked.
Therefore, when expanding the multi-level content, please be sure to clarify the object.
For example, replace [please introduce the second point specifically] with a more specific description such as [please introduce the second point in the outline], so that the AI will not confuse you.
Topic 2: Pay attention to the relevance of context
As we mentioned earlier, AI has a powerful multi-turn dialogue and the ability to relate to context.
If we intersperse multiple different topic scenarios within the same dialog, then the AI's answer may be affected by the previous content and appear indiscriminate.
So, in the interaction with GPT, if we want to ask multiple different topics in a dialog box, then I recommend that you initialize GPT when a new topic is opened.
That is, clear the previous conversation before starting a new topic, so that the AI's answer can be avoided from being disturbed by the previous content.
The specific operation to reset the ChatGPT prompt is as follows:
Well, here we have finished the [Continue] instruction for tuning AI~
Although the instruction can make the AI's answer more colorful, but subject to the limitations of language transmission information, each answer of the AI may not always be as we want, and even chat and have the phenomenon of deviation.
In this way, there will be a lot of unnecessary trouble when we get the desired effect.
Then in this case, we need to use the second instruction of tuning AI, which can help you design tasks with "routine properties" and "template classes" with magical effects.
2. Adjust the reward and punishment instructions of GPT
In fact, the role of this so-called "reward and punishment instruction (supervised learning)" and the principle of tuning are well understood, it is like we educate children.
If you want your child to meet your ideal standard of behavior, then you need to educate him, and if the child does a good job, we need timely praise to encourage him to become better.
And if the child does poorly, we need to punish him and let the child know your bottom line.
So through such repeated correction of mallet + reward, children will naturally form a set of behavior standards that we expect.
Therefore, since the AI based on neural networks has the ability to think chains, then of course we can also follow the idea of educating children to make AI more and more obedient.
As for the practical idea of this reward and punishment instruction, it is very simple, here you may wish to take the [problem assistant] I trained as an example~
This should be known to my students in the growing circle, and every week I will give a thinking question to my community~
So, after the advent of ChatGPT, I was thinking about whether I could let AI automatically or assist me in generating thinking questions to share this part of my workload.
So, under the guidance of this goal, I started the tuning of AI....
The idea of tuning is simple:
First, feed me the previous thinking questions, then let it learn and analyze the format of my questions, and then let it generate answer patterns.
For these AI-generated content, if it comes across something that meets my requirements, I will use affirmative words like: Very good, please keep it in this form.
For places that don't meet my requirements, I'll use negative words, such as: No, you're wrong, please re-ask for XXX.
(Normally GPT will apologize to you right away and correct its mistakes)
Similar to what you see.....
So after repeated feeding and round of tuning, it finally became a small assistant to assist me in solving the problem~
Well, if I don't tell you in advance, can you tell which one is an AI problem...
Therefore, you may wish to think about the scenarios in your work to see what are the more patterned tasks, for these more patterned goals, you can train the AI through [reward and punishment instructions] ~
In the process of training, encourage those that meet your requirements and punish those that do not.
Finally, through such back-and-forth rewards and punishments, the AI will meet your ideal behavior standards, generate your ability to meet the content you want, and become a long-term assistant in a certain scenario in your work.
Well, here we have finished talking about the method of [training AI technology]~
But we also know that knowledge, technology, and tools are ultimately designed to help us create actual viability.
If you can't achieve this goal, then even if you master the dragon slaying technique, but there is no dragon to slaughter, then this dragon slaying technique has nothing to do with you.
The core reason why many of our students got GPT and mastered the use of GPT, but still used it very often, or even couldn't use it at all, is because you don't associate AI with the application scenarios you need.
Therefore, if you want AI to create actual productivity for you, you must find your own application scenarios, and only after integrating AI into specific scenarios, our dragon slaying knife will be useful.
So in the next part, I will share with you, on the basis of mastering the operational AI methodology, how to combine AI with yourself to bring actual productivity~
3
How to integrate AI to create real value?
In fact, the idea of integrating AI and ourselves is very simple, summarizing the whole idea, which is nothing more than two steps.
The first step, I call it: combing
In fact, this [combing] is very easy to understand~
When e-commerce was rising, Ma Yun once said that all business is worth doing again with the Internet.
Well, in the era of AI, I want to paraphrase the same thing: almost all ways of working involving knowledge can be reconstructed again with AI.
Other words:
We can sort out those parts that may be replaced or assisted by AI by reflecting on our own work scenarios, and then find the points that they can combine with AI, and create standardized tools or processes according to the ideas explained earlier.
To summarize it in human terms: sort out all your daily work trajectories, find the work that AI can do, and then give it all the things that AI can do or let it assist you to do.
So as to free you to do more valuable and creative things.
Well, here we know the connotation of [combing], so how to do this combing?
In fact, this idea is very simple~
You can focus on the universal three dimensions of your life fundamentals (study, work, and life), and sort out the scenes in each dimension one by one according to our behavioral needs.
Through this kind of systematic thinking, all the scenarios that can be combined with AI are caught.
For example, in my learning scenario, there are these scenarios where AI can replace or assist....
◎Learning scenario case 1: Use AI to assist in processing knowledge, explaining knowledge, and providing inspiration.
◎Learning scenario case 2: Use AI to achieve conversational learning with masters
Of course, in this scenario, in addition to the dialogue with Lao Tzu, you transfer your thinking and replace it with Confucius, Zhuangzi, Mencius, Mao Zedong, Napoleon, Socrates....
In short, any great man you are interested in and you want to talk to.
Even let GPT play multiple roles, let the big guy and the big guy duel each other, you come to watch the battle, learn from the dialogue, let the learning return to the "Socratic state".....
◎Learning scenario case 3: Use AI to achieve assisted reading and improve comprehension efficiency.
This application scenario, for those students who do not have strong comprehension ability, or want to improve the efficiency of understanding, is simply an artifact, such as.....
◎Learning scenario case 4: Use AI to achieve entry coaches and mentors in major fields.
This scene is simply a gospel for those students who want to build a certain field system, or want to systematically study a certain field.
For example, this case of technical learning.....
Moreover, by the same token, in addition to this "technical" field, you can also transfer this idea out and use the same routine to study philosophy, sociology, product managers, operations....
In short, any field you want to study, you can let AI show you the way.
Of course, in the learning scenario, in addition to these, there are N more scenarios, such as, be my English coach, debate coach, learning effect tester, etc.....
Due to the reason of the space, I will not give examples one by one, if you are interested in more usage of [learning scenarios], I will write an article another day to do a separate topic sharing.
Well, here is a brief introduction to the combing and demonstration in the learning scenario, then the same routine is also true in the work scene ~
◎Work scenario case 1: Use AI as your job assistant.
There are many examples of this assistant scenario, you can use it to help you write job postings, write scripts, write job copy, self-media copywriting, write code, and even write programs.....
For example, I asked AI to help me write this product manager's PRD document, after reading it, do you think that this part of the work is already in jeopardy in the future....
Similarly, use AI to do data analysis....
Or use AI to write meeting invitations....
Of course, in addition to the above more professional scene assistants, it can also be fully qualified for creative or stimulating creative work.
For example, the title of our article is inspired by GPT....
After reading this, can you suddenly understand why after ChatGPT, many companies began to lay off employees....
In short, use our above prompts and training routines, then everything can help.
Of course, in addition to these, there are N multi-uses in the work scenario, such as, let it be my typesetting & topic selection tool person, screening and integration data engineer, etc....
In short, the examples under the work scene are also endless, if you are interested in the work scene, you can follow the above number to add me.
Find me for free, more scene cases to help you broaden your thinking, and some mirror websites that can be directly accessed in China.
Because of the length of the article, we will not give examples here one by one~
Well, here is the end of the first two scenes, then the same, in our life scenes also operate ~
◎Life scenario case 1: Use AI as a fitness trainer~
Another example, use AI to be my personal nutritionist....
Of course, in addition to the above examples, the use of AI in our life scenarios is also endless.
For example: let it be your personal lawyer, personal doctor, private tour guide, etc., all kinds of coaches, all kinds of identities, it can be done very well!!!
In short, whether it is work, study, or life, there are countless examples and scenarios like this, and I can't give you an end, and you can't finish learning.
Here are these scene cases, just for you to broaden your ideas and inspire use, the examples themselves are not important, the mining ideas behind the example scenes are what you should really learn.
As long as you can systematically sort out yourself around [your three-dimensional daily trajectory], in the process of combing, when each scene jumps out, think about the point where the scene can be combined with AI, record it, skip if not....
As long as you carefully complete this process, then you will definitely be able to dig out a large number of useful scenes, as well as unique usage, and these scenes you have excavated with your own hands are the scenes you really need and can solve the problem for you immediately! ! !
Well, when we dig out the scenes that can be combined with AI through the above combing, we can bless these sorted scenes with AI to create productivity, which is our second step~
2. Landing GPT Step 2: Build
In fact, this so-called "building" refers to: to sort out these scenes, those scenes that can be standardized and reused, so that they are all standardized and instrumentalized to form a "scene library" for us to call directly when we need it.
For example, the thinking problem assistant I trained earlier, the fitness trainer assistant and so on....
As for why you want to do this action, it's also very simple~
That is to say, unless we have a certain degree of visibility and the data has the value of being captured by AI vendors, the data trained by us individuals cannot enter the training set database of AI manufacturers.
That is to say: even if you train the AI in a dialog box very obediently, when you reopen a dialog box, then all the interaction memories between the AI and us will disappear...
Therefore, for some valuable scenarios, we need to save the [scene data] we have worked hard to train, so that it can provide us with long-term services, rather than a one-time transaction. For example, take the [mail assistant] I trained as an example~
Every time I need to send an email, I just have to tell me the content of the email.
And for the mail, salutation, format, writing style, etc., I don't need to tell it repeatedly, it will automatically read the previous data....
Similarly, similar to fitness trainers, nutritionists or other scenarios that require repeated interaction, etc., this is the case~
As long as you train this scenario, then the next time you need it to customize the scheme for you, it will automatically read the data in front of you and combine your previous situation to give you the best solution at the moment
After all the interactions, you don't need to explain and supplement a lot of background information on it.....
Therefore, when we save these trained scenes according to certain rules to form a scene library, we can repeatedly apply them to help us save a lot of time and energy.
Well, here you know, after the meaning of creating a "scene library", how to operate it specifically?
Then I will share two ways with you here~
Method 1: Use GPT's built-in scene library directly
In other words, every time you start a new conversation in GPT, GPT will automatically create a dialog (scene) for you.
Then we can leave those scenes that have training value and can be fixed, and then classify and name them in our three-dimensional way.
For example, an effect like this...
Of course, the limitations of building a scene library in this way are relatively large~
Because the official one is in a fixed format, we can't classify and sort our scenes under the limitations of this method.
In this rigid format, once you have too many topics in the conversation, then this troublesome energy of rummaging back and forth will have a feeling that makes you want to hit the wall.
Therefore, compared to this way, I recommend you the second ~
That is: in your external brain system (note management software), with a more flexible central control page, they are centrally managed, and the effect is similar to this...
In fact, the implementation principle of this method is very simple~
If you look closely, you will find that each dialog (scene) in GPT has a separate independent URL~
Therefore, under this mechanism, we can take down the link to the scene we need, and then classify it in our external brain according to the logic we need.
In this way, when we call each scene, we don't need to open the GPT website, and then toss and rummage for half a day to start~
Instead, go directly to our center console, click on the corresponding link, and it will automatically jump to our corresponding scene dialog immediately.....
The whole operation process will make you silky have a feeling of flying~
Then, as your AI landing scenes increase more and more, you will form a set of completely your own [AI scene library] in this panel, what you want, everything is within reach...
Well, at this point, our entire AI user guide is basically over~
In short, when you can follow our three-dimensional thinking to do systematic combing, and then, according to the writing skills and training AI methods mentioned at the beginning of our article, train it.
Then, after pinning them into your AI scene library, congratulations, you've successfully hired a free assistant.
It will help you share a lot of trivial tasks in your various scenarios, and help you fully realize the great boost of work, life, and study efficiency.
As long as you do this seriously and arrange it in place, not to mention increasing your efficiency by ten times or eight times, then increasing the efficiency by 2-3 times is more than enough.
Of course, I have to remind you a little at the end~
AI has not evolved to the extent of going against the sky at this stage, so in many scenarios, it still has no way to achieve the effect of real people, and the answers it generates are not 100% correct, and in many cases it will also have fabricated answers.
Therefore, for the content of some important scenes, we also need to manually polish and modify it, and trace and check the information it gives.
In short, even with the assistance of AI, do not lose the ability to think independently, whether it is now or in the future, people who think independently can control AI, not be controlled by AI~
4
The whole context map of the article~