ChatGPT-4 is so effective, will many programmers lose their jobs?

The cover is "Detroit - Become Human", a game about robot AI that I personally like very much.

This is a question I saw on Zhihu, and there were hundreds of answers in just a few days. In addition to proving the current top status of ChatGPT, it also reflects some problems. There are some factors of anxiety or panic among the general public.

In fact, it’s not just the general public. If you pay attention to the news, you will find that there are already many voices in the United States calling attention to the hidden dangers brought by AI, and even stopping the training of ChatGPT-5. Regarding this matter, I actually expected it. The white left is rampant in the Western world. Once certain industries make a slight improvement, they will often be labeled as anti-monopoly, human rights investigation, environmental protection issues, etc. I am not saying that these are not important, but development is the last word and these should not be ranked in front of development.

AI’s influence and popularity are so great, there is no reason why it should be an exception. Of course, this is a good thing for China. It is best to call a halt, which just gives us time to catch up. Let’s go a step further and let’s get back to the problem itself.

I personally recharged ChatGPT-4 plus. After using it heavily for several days, it was no longer as amazing as it was at first. Especially after recently replenishing knowledge on the relevant principles of large models, from a principle point of view, large models are also models based on probability and statistics. Essentially, the goal of model learning is to generate the answer text with the highest probability and best effect based on the input text. This is also the reason why ChatGPT3.5 version often talks nonsense seriously. Our answers that seem to be nonsense are exactly the most probable results in the eyes of AI, such as the classic parents’ wedding question.

The same goes for writing code. ChatGPT code is also generated based on the same logic. It has the ability to understand and edit code to a certain extent, but because it has not actually undergone actual programming work, many times the answers are a bit speculative, especially for some relatively niche questions that do not have ready-made answers. When I was developing tools based on the ChatGPT api before, I often encountered the given code that was specious, and there were some problems like that.

So if ChatGPT continues to iterate and achieves further results, can it replace engineers? We make an optimistic assumption and assume that GPT-4 has an average level of engineers with 2-3 years of experience in various coding fields. So, can it replace programmers?

If we just look at AI-related demonstrations, it is normal to have worries and anxiety. Sometimes I wonder, if the ChatGPT code is written more smoothly, can it replace me? But in the blink of an eye, there was a review meeting the next day, and I felt relieved instantly. ChatGPT can write code, but can it still report to the boss?

So the answer is yes, no.

I believe that anyone who has had some actual development experience should know that program development is not just about writing code. Let alone writing code, even if it is writing a media article, no media company would dare to let GPT write the article and then publish it without review, right? Any editor who dares to do this will probably end his career.

The same goes for writing code. It doesn’t mean that the code is just written and put there. Before development, a requirements review meeting should be held to understand the specific content and various details of the review requirements to ensure that they are fully understood. During the development process, progress meetings are also held with other upstream and downstream collaborators to discuss and formulate various specifications such as API interfaces and data, and to coordinate the progress of all parties. After the development is completed, joint debugging and testing are required to ensure that the entire upstream and downstream links are normally smooth. After the test is completed, the product manager and the requester must be invited to review to ensure that the output results meet the requirements, and finally the product is released online.

Especially for some large companies, the entire development process is very complicated with various rules and regulations without talking about code. Especially things with huge impact like going online require layer upon layer of approvals. Many times when various problems are encountered and blocked, people are definitely needed to communicate and coordinate.

When an engineer completes a requirement, he may only spend 20% of the time actually writing code. Most of the remaining time is spent in meetings, testing, debugging, etc. In other words, I don’t spend much time dealing with computers and codes. Most of the time I deal with people. After all, code is written for people to use. No matter how beautifully written the code is, it has no value.

Secondly, although GPT-4’s context understanding ability and coding ability are very good, and even reach the level of excellent engineers in some cases, there is no way to replace humans, even for some so-called most basic additions, deletions and modifications. Check the engineer.

The reason is also very simple, because in actual projects and development, there is a lot of background information. Speaking of human language, there are too many patches and pitfalls, which is what we often call word of mouth. In fact, what is passed down by word of mouth is some key information in the development process. This information is very complicated and is not included in the documents. Some of them are also difficult to describe in documents (for example, a certain requirement was proposed by the boss, and a certain bad design was personally proposed by the boss. made). It all relies on engineers passing on from person to person and accumulating experience over time. It will take a month or two for new senior human engineers to become familiar with it, let alone GPT.

Should I ask my boss to write a huge document every time there is a need to document the pitfalls and historical issues in the project and throw them to GPT? Not to mention how long these documents are for some complex projects. Even if it is actually written, what should I do if there is a problem with the written code because GPT does not understand it? Wouldn't it take more time to communicate with GPT repeatedly and ask it to make changes? The boss chatted with GPT when he didn't do other things all day, right...

The boss must have had the time to argue with GPT and get it to be revised. I had already finished writing it myself.

Finally, I will briefly share my personal understanding. GPT-4 can write code and answer some questions. It is actually a very powerful and effective tool. The significance of the birth of tools is to increase productivity, liberate human labor, and allow humans to devote their energy to other higher-level work rather than simply replacing humans. Simply put, using GPT can improve programmers' coding efficiency, and the same people can do more work.

On the surface, it seems that a group of people will lose their jobs due to the improvement in efficiency. But this is based on the premise that demand is constant, but in reality this is not the case. The development of technology does not just eliminate old productivity. It can reduce the cost of use. Some things that could not be done or were expensive before are now cheap. , then it is very likely that new needs will be tapped and new industries and opportunities will be brought about.

In the era when there were only horse-drawn carriages, only a handful of people could travel far. Although the popularity of automobiles seems to have eliminated horse-drawn carriage drivers, it has also made more people willing to travel far, directly igniting the demand for travel. As a result, a series of industrial chains such as petroleum, hotels, and highways have been developed around automobile travel, thus creating more jobs. This is also the reason why the proportion of farmers has been declining rapidly since the Industrial Revolution. Because productivity has increased, 20% of farmers are enough to meet everyone's food needs, and others can do higher-value work.

From this perspective, GPT is more like a car in the new era. It removes the burden from the carriage and therefore allows humans to go further.

The above is the detailed content of ChatGPT-4 is so effective, will many programmers lose their jobs?. For more information, please follow other related articles on the PHP Chinese website!