Datawhale dry goods
Author: Ai Fu, Fudan University, Director of People's AI Education
foreword
Good evening everyone, I am Aifu, the person in charge of AI education at Minmin AI.
Let me introduce myself first. I am a middle-aged slasher who repeatedly jumps in the technology circle and the education circle. About 5 years ago, as a technical person who had written code for many years, I built a bridge between technical engineers and information technology teachers in primary and secondary schools. The vision at the time was to take children to understand this computer-adapted world, so that everyone can make children powerful because of computers.
Then at the beginning of this year, it was very obvious that not only children, but adults also need to actively face this wave of AI. Everyone also needs to learn to talk to AI, and then Learn to use AI well, and be powerful because of AI. So I returned to the common people as the person in charge of AI education, and now I have been exploring how to help every ordinary person use AI well. We are not only for computer majors, but also non-computer majors or non-students who have already worked. Indeed, everyone needs to actively embrace this wave of AI. I am very happy to do a sharing. This sharing will focus on the actual use of AI, and what kind of abilities need to be cultivated in the future career development. I hope it can give you some inspiration.
This picture is quite interesting, the engineers who developed ChatGPT. Because ChatGPT can kill many things, he also killed other engineers, but ChatGPT can also write code at the same time, so he also killed the engineer who developed ChatGPT. In fact, is this really the case? Let's look at KentBack, the initiator of XP Extreme Programming and the leader of agile software development two days ago. He sent a Twitter saying that he had been reluctant to try ChatGPT before, and finally overcame his reluctance. Afterwards, he found out and understood why he was reluctant before, because 90% of the skills had been reset to zero by ChatGPT , and the remaining 10% leverage had increased by 1000 times. The first half sentence is bad news, but the second half sentence is absolutely good news.
At this time, we need to consider this question. In the era of AI, what is your ability to be magnified 1000 times? In fact, in the past period of time, because of the in-depth cooperation and exchanges with many innovative education teachers in the education circle, everyone has been discussing this issue recently. What should students learn in the AI era? What is the teacher going to teach? Now that AI can help students cheat, help students pass various exams, and complete various homework, should we as teachers guard against it, or should we think about it the other way around, whether these things do not need to be taught? , no test is required, we should teach some abilities that are more needed in the AI era.
Core competencies magnified a thousand times by AI
What will these capabilities be in the AI era? Let me give a small example from a programmer's perspective:
This May Day, because I got the code interpreter of ChatGPT, its waitlist finally released me, that is, you give ChatGPT a line of command, a command, and then it writes a program, which can automatically execute the program. So I had a lot of fun this May Day. I first did a small experiment and asked GPT to help me write a Tic-tac-toe game. I set up two players, and then they started automatically after I said start Play the game. Sure enough, GPT quickly wrote out the entire program, and it can run normally.
It is a very simple sentence, and it can be understood and done well. Does that mean that the ability to code like this has been reset to zero? In fact, let's look back. After he helped me complete the program of this ability, I found that all its codes are contained in a class. Although the function of one line is clearly written, the whole design is still lacking. . And the lack of this part was indeed reflected when I asked him to make a small change next.
I said at the time, "This chessboard of yours is too ugly to draw, and you need to adjust the format of the chessboard a little bit." As a result, he changed it repeatedly, changing one place each time, and then reported errors elsewhere because the entire code was lumped together. So he apologized to me more politely, saying I'm sorry, I forgot to modify something, so he modified it and ran it again, and reported an error again.
At this time, we realized that GPT is not capable of making a good design, because after all its codes are grouped together, a slight adjustment to the function will lead to a large number of errors, which will eventually get out of hand. At this time, the 10% capacity mentioned by Cent is needed. When humans come on stage, I reorganized the language and gave GPT a very clear instruction :
In the same way, being able to realize such a game of playing Tic-tac-toe does not seem to be very clever. The role played by humans here does not seem to show value, but we can see the power in the next step.
Because after having a good design in front, GPT has successfully completed the required changes every time. Then the game continues, and the next thing that is really powerful is what is framed in the red box above. I gave it a chain of thought and said that please follow the steps below to create a new player:
The first step is to analyze the reasons for the failure of the last round and think about whether there is a better strategy to win the game.
The second step is to adopt a new strategy, write a smart go five player, and write a new player class.
The third step is to let the new players compete with the winner of the previous round for 10 rounds and see the results
And repeat the above three steps for 3 rounds to see if you can create more and more intelligent players.
Sure enough, because GPT had a good design, it implemented it step by step smoothly, constantly thinking about whether there is a better strategy, and then writing it out for execution. In the end, he even thought of using Minimax, Alpha-beta pruning and other algorithms, which are very complicated algorithms.
At this time, you can see a powerful combination of humans and AI. For humans, the role they play here is to help achieve a clear architecture design, which can be well adapted to subsequent changes. AI can do everything for me. Although it's not like the 1000 times magnification that Cent said, there are always dozens of times.
If I really want to iteratively write, then three or five times the algorithm, I guess it will take an hour, two hours, but AI really only takes a few minutes to get it done. Then, the combination of human thought design and AI execution, the powerful combination here allows AI to iterate left and right to create more and more intelligent backgammon games.
This is a small example. From the perspective of a programmer to think about where the competitiveness is in the AI era , what we can see is that those in the industry that require in-depth knowledge, skills, and experience are the ones who will Being magnified may not be 1000 times, but it will be a very large 10% magnification.
This means that for programmers, it may be an architecture design ability , and it is a dismantling plan for a field when facing more complex fields . Then maybe it is low-level for legal personnel, and the low-level legal documents that can be replaced by AI are those low-level legal documents, but there must be very high-level thinking of lawyers that will not be replaced.
So everyone is in the classroom, usually in the classroom of school, or at work. In fact, it is possible. You should have some feelings like this. You can clearly describe which things are repetitive and mechanical. Give some of the things that execute the logic to the AI , and report the work to the AI, so you can relax. And others that require you to think harder. You need you to have more in-depth knowledge and skills, and these will be magnified for you.
There is also a small tip, which is to check whether you have fully used AI. This is a conversation I had with a mathematician friend a while ago. He was very interesting. He said that after using AI, he was even more tired. why? Because he has been doing a very grand mathematical theory research in the past two years, and then he needs to have a lot of proofs, and then the proof of the small step from a to b, maybe he needs to check some that he himself does not know. Familiar formulas, and then step by step to complete this short period of proof process, it may be a regular and mechanical labor in one or two hours. But now he handed over such mathematical reasoning to GPT, and then what he wanted to do, he continued to do the original one or two hours of work, and GPT helped him complete it in 40 minutes.
He needs to keep thinking about what to do from b to c after the reasoning from a to b is completed, and where is d after c? He needs to keep thinking about these things, and the density of thinking increases. Then I used to work all morning without a break, and I could do it all morning with special concentration. But now basically because the density of thinking in this way has increased, because it is definitely more difficult for you to think about higher-level things such as a to b and b to c than to think about the step-by-step reasoning of a to b. Because after thinking about the increase in density in this way, I feel that I must take a break for more than an hour, so I also teach you this skill, that is, you can make full use of yourself according to yourself. In the process of using AI, does the brain feel Even more tired, thinking about whether the information density is higher, to judge whether you have really made good use of AI.
Deliberate practice to improve professionalism with AI
Let's give some simple examples first, let AI help write resumes. Of course, everyone knows that AI can help you write a lot of things, but if you tell it that I am a Python backend and write a resume for me, the result you get must be unusable. This is a wrong posture:
Is this next one better? It does compile a resume. Is this resume available? The first is that even if you get lucky enough to pass the resume screening and get an interview opportunity, you will probably be in a lot of trouble, right? Because it is edited after all, there are many things in it that you may not be familiar with. Second, in fact, as someone who has done a lot of recruiting and seen a lot of resumes, I can honestly say that resumes like this usually don't pass the screening. Why? Because all these words of familiarity, fluency, understanding, proficiency, etc. are too thin and empty, it seems that you are more referring to the needs of my ladder, and then pile up this pile of things That's all.
I don't believe anything in it, so what should I do to write a resume that can impress the interviewer and catch the eyes of the recruiter? We need to analyze the content of this JD and this job description in depth. We have to read a few pieces of information from it. First, which experiences and skills recruiters pay attention to, and then consider whether their experience, skills and job requirements match in a targeted manner. Second, if there are matches, how can the resume be written to highlight these matches? Especially how to write to highlight the details of these matching items, and use the details to catch the recruiter's attention, then let's look at it step by step.
The first is to let AI help you sort out the needs of JD.
I ask here, according to the checklist above, you should give me a score according to its category and its specific requirements, and give me a weight of 1-5 points according to the job responsibilities, and help I marked whether it is suitable. Whether it is appropriate to leave a blank column, let me tick it myself. In order to avoid GPT forgetting, I posted the job responsibilities here again, because the previous text is already relatively long.
Finally, there is a little trick. The output of Markdown format is a colon. With this, we call it match boost. With this boost, you can effectively avoid GPT talking nonsense, and it starts to paint tables in Markdown format for you. So this is a pretty useful little trick.
By the way, what we can see is that after having such a Prompt, GPT really listed a table for you in an orderly manner, listing each specific requirement in it clearly, and also made a score.
Ok, we saw that these two steps have sorted out the requirements of the ladder, but what do we need in the next step? We need to dig deeper into why JD has such a request? So the prompt I give here is to analyze the responsibilities of each position. Here I have made such a prompt for each ability.
Through such an analysis, one thing we can do is help to further understand why the recruiter listed these capabilities in his node. What does he want you to do in the past? In fact, he also helps you to check whether this job is what I want, whether I want to do it or not?
First, am I capable of doing this? Second, do I really want to do this?
If the answers to these two questions are yes, then we can further think about how to write a resume, how to write to reflect that we have these abilities, and more importantly, to reflect that we have these abilities through the introduction of project experience. these abilities.
Therefore, after using the previous methods to let GPT sort out for yourself, you can actually list your abilities more specifically in the introduction of the entire resume. To summarize, we can use GPT to help:
The first step is to sort out the requirements of the ladder.
The second step is to understand what the recruiters are focusing on based on the ladder requirements and job responsibilities.
The third step is to write a resume based on the key points of concern and combine your own project experience
The fourth step is to delete, adjust, beautify, etc. according to the content spit out by GPT
The fifth step, I hope to be able to say that the resume must be consistent with the person, even if you are still a little bit short, it is best to try to make up for it before the interview, right?
Okay, this is a small example of the use of resumes. Suppose now that with the help of AI, you successfully passed the interview and found a job you like.
Next, let's talk formally about professionalism. Two days ago, I happened to be chatting with a business school teacher about the possibility of AI helping professionals in workplace writing. At that time, the teacher of the business school was particularly touched, because he came into contact with so many business managers every day and heard the complaints:
In fact, we can really regard AI as our own personal career coach and career development coach. Why? Because AI is actually a weighted average of all human knowledge, it does possess the general knowledge of all human beings.
Then how to do something with professionalism in the workplace. Let's look at some small examples, like here where AI helps improve a business email, here I will flatter AI and hypnotize him:
After these 4 steps of thinking, let's talk about actions to improve the current email, and the last paragraph is particularly important. If there is any information that needs to be supplemented, you can use placeholder to replace it first.
So with the guidance of such a chain of thinking, attach the content you want to express: "We have an event on the weekend, can you provide the stage to build lighting, sound, tables and chairs?"
After you have such an email as the base, it will become very easy for you to make some modifications on this basis.
To sum up, if AI is used to help improve business emails, we need to clearly explain the background, make requirements, and then use the chain of thought to improve the quality of email responses. Then there is a little trick for some tasks like this, you can try to regenerate several times, because the AI will do it every time, that is, unless you let it spit out 20 things that need attention, etc., otherwise the AI will do it It's almost enough to spit out three or five, so sometimes some things will be missed. Then you try a few more times, you may find some different details each time, and then make a summary yourself.
You can also make a modification, which is to let the AI output its own thinking process. This kind of thinking process can actually help you understand the main points of writing emails, and then help you to better infer other things from one instance and integrate them later.
Here we take the meeting minutes as an example. Then you can see that the minutes of the meeting are indeed very simple, that is, they diligently recorded some key points that everyone said at the meeting, but recorded them one by one in chronological order. Then, because the meeting was very random, everyone may discuss in a messy manner, so the entire minutes are also very messy.
So what to say in this case? Tell GPT, please follow the common template of the excellent meeting minutes of the product project team of the Internet production company. And what exactly are the template requirements for these things? Needless to say, he was asked to improve the minutes of the above meeting because he naturally knew about GPT. For the part that lacks information, you can write the content as a demonstration, and mark it with a box, so that I can adjust it according to the actual situation. This one is the same as the previous one, because if you don’t add this sentence, he won’t make it up for you, and then you may miss many important details. Finally, we added this sentence, please explain the main points of each item of revision, so that I can write better meeting minutes in the future.
OK, you can see the minutes after the improvement with the help of AI, and then see that the format needs to be recorded by the mother to the group to record the time and place of the minutes. The host, recorder, and participants, he even edited each role for you. It deduces what role a person is, and then compiles it. Then the most important thing is that so many messy seven or eight messy pieces of information, after being understood by him, were sorted into three important items, and then all of them were arranged in the content of the meeting, one by one. The appropriate location, and what is the conclusion of each subsequent meeting, and what is the next step, it also lists them for you.
A square bracket was added to the next action because it didn't know the time, so it made up a time for you. On the other hand, in fact, if you understand why AI is written like this, it can also help you become a better professional person.
So everyone should have a feeling after watching such an example, that is, AI can really play the role of a career development coach, and then help you practice deliberately during a conversation with him, and integrate all the content needed in the workplace. Turn it into your knowledge and skills and form a habit. Of course, I feel that after I have introduced these, I don’t know if some friends will say that AI can do these things, so why should I practice deliberately? Why do I still need to know it? Anyway, if the boss assigns me this task, I will just let the AI do it for me. It’s over. AI knows it, so why do I still need to know it?
Here I want to talk about an earlier example. In 2008, there was a particularly famous article, "Google Making Us Stupid?" written by Nicholas G. Carr. "At that time, the Google search engine was indeed very popular, and everyone was used to finding answers from Google.
Nicholas asked, I noticed that in the past, in order to understand a piece of knowledge, my child would flip back and forth on the Encyclopedia Britannica to find the answer, and during the flipping process, he may have also received a very large amount of related knowledge , but now he Googles it, and it ends when he sees a one-sentence answer. Knowledge is too easy to obtain. After the friction is reduced, people's curiosity is damaged, reducing curiosity, and it even makes people have some illusions of knowledge. I thought I knew the so-called truth, knowing so much would not lead a good life, but it was the same. We cannot deny that a search engine like Google is undoubtedly a huge amplifier for those who make good use of it, allowing you to obtain knowledge anytime, anywhere when you need it.
But for lazy people, in fact, it will make people stupid. The same is true for AI. AI helps us make many things easier, but in fact, there are some things that really have to be paid, and there must be hardship. obtained after the process. All these detours that you have gone through are what you must go through in your growth and the nutrient for your growth. So there are some things that are actually indispensable, and only after some deliberate practice will they be internalized into your skills. Otherwise, if one day you suddenly cannot access AI, it may be very confusing.
Then just two days ago, I chatted with my friends about a particularly interesting analogy, and I also shared it with you. That is, we said that when the industrial revolution was liberated, everyone’s physical strength has indeed become easier, but what we can see, obesity and chronic diseases, become the biggest killer of human health, so it is inevitable that there will be such a situation a worry. After the AI revolution liberates the brain, we don’t need to think. After many things don’t need to be done by ourselves, will mental illness become the new killer of human beings?
This is a very serious problem. We are actively embracing AI, and we believe it can bring us great prosperity and a burst of creativity. But with so many examples listed above, it feels like there are times when I might be able to lie in the metaverse and get drunk. Under such circumstances, the torture of why we human beings are actually something that everyone needs to think more about.
Explore new possibilities of unlimited practice with the help of AI
What I just mentioned is about the workplace, that is, what to do when you are a worker. I also know that many young friends today hope that they can work in a more free way, whether they are freelancers or their own bosses. In fact, after the emergence of AI, many, many people regard AI as their own outsourced employees, outsource all work to AI, and then become a super individual themselves. Such things have become possible.
So our third part is going to talk about using AI to explore a new possibility of unlimited practice. The reason why we want to explore is also very simple. As mentioned earlier, only a boss like Cent may have the ability to magnify his own ability by 1000 times with the help of AI. For many newcomers in the workplace, for many students, it may be magnified by AI 2 times and 10 times.
It is like the picture on the left, the ability picture, we can see some of my abilities, some will be zeroed by AI in the future, and some will be doubled by it.
Then with AI, if I only focus on my original abilities and my original identity, I am a programmer, I am a designer, and I only focus on these abilities that I already have, it will It became very narrow, although it was a little longer, but very narrow. But we still have a possibility to turn many 0s that we didn’t have at all into 1s. We can do a little bit of these abilities that we didn’t have before, and we can do a lot more things with the help of AI. , can do a little bit of things.
A particularly simple example, data analysis. In the past, this may have to be an engineer, a specialist in data analysis, or at least be able to write SQL to be able to do it. In other words, although there are a large number of data analysis tools now, it is still somewhat troublesome to use. But after having AI, this is also an experiment I did on the code interpreter, I passed it an Excel file, and then told him to help me analyze all these video content, each person’s effective viewing time, and then follow the video The title of the video is aggregated and placed on a picture one by one. For example, AI directly writes the code to execute, and presents a histogram for me, saying that the sum of the histogram is not very make sense, and it is better to analyze it with a quadrant diagram, so he showed a quadrant diagram and gave him a line of commands, saying that it is vertical Colored, a colored quadrant chart, in just a few words, all the charts I need for analysis will appear.
what does that mean? This means that a person who does not have coding ability or data analysis ability, but may have good business intuition, good common sense, and know how to make some qualitative judgments from common sense, he can now rely on GPT carried out some quantitative analysis. Data analysis has become a very common skill.
What is more common is that I have just seen that painting AIGC has indeed enabled many people like me who do not understand painting at all to be able to create something, such as the one in the middle. The content is designed by a teacher who is going to give an AIGC SD drawing class. You hold the drawing and simply draw it by hand. It can generate pictures of different styles for you. If you take these words, it can generate some pictures for you, and if you take some words, it can generate a 3D logo for you.
The same is true for programs. How many good ideas we have before, we only need a programmer, right? What about at this stage? At this stage, no matter for programmers or non-programmers, you can actually think about it. AI is here, what can you try to do that you have never done before?
Summarize
Later, a former programmer said, you really can't empathize with my anxiety, I'm still worried about losing my job, have you ever thought about it? But one of our worries at the time was, have you ever thought about it, although many people can use AI to complete the tasks of programmers, but as a programmer, what are you interested in? In fact, you are fully capable of grasping the fur of some other majors, whether it is law, biology, anthropology, sociology, all of which have become much easier.
So, are you really willing to repeat some mechanical and boring code? Or do you prefer to conceive some things, move your lips, and let AI help you magnify your capabilities by 1000 times , and directly complete these things. So I believe that we can do more creations in the future.
It's not easy to organize, so I like it three times ↓