OpenAI's new model ChatGPT has only been born a few days and has become a "force weapon" for the majority of netizens.

ChatGPT, which launched last Wednesday, has exceeded the 1 million mark in just a few days. "the CEO Sam Altman representation of OpenAI.

Whether its answer is correct or not, it looks like a bunker at first glance.

In addition, it is very easy to generate, and for a while these "looks good" answers are flooded to the legendary Q & A site running with codes around the world, the real teacher of computer students-Stack Overflow.

However, in order to judge whether these answers are right or wrong, a large number of people with professional knowledge need to examine them carefully.

Forced to have no choice, Stack Overflow had to issue a declaration: ChatGPT, banned!

There are so many mistakes that they should be sealed.

With the heated discussion, the official side of Stack Overflow, the second largest programmer dating site in the world, is also clear-cut:

Do not think about using AI robots to confuse the public, send a good answer, do not use ChatGPT to directly generate spam answers, otherwise the title will be sealed.

This is a temporary rule in response to the current flood of ChatGPT-generated content on the forum, Stack Overflow said in an official notice.

The circular points out that the main reason for this rule is that ChatGPT automatically generates answers of low quality, has too many mistakes, and looks like that, even people who don't understand it can generate answers casually.

As a community based on the quality of Q & A, these low-quality answers can cause serious trouble and inconvenience to those who come here for help, as well as harmful to Stack Overflow itself.

Stack Overflow said it would continue to weigh new policies around ChatGPT "and other similar tools" in the future, but the date of the decision and the content of the final restrictions were unclear.

In short, the attitude of the Stack Overflow side is very clear, to sum up: "play back to play, play back to play, do not joke about questions and answers."

Netizens clapped and applauded

As for this decision, the onlookers at the bottom fully agreed with it.

-- well done! I really hope that this decision is not temporary and is not limited to ChatGPT, but should be extended to all AI-generated answers. AI will never give a high-quality programming answer, and it won't happen in another 100 years.

This ban should not be a temporary measure.

Using ChatGPT, you can brush points through crazy irrigation and automatically generate answers, even if they don't know and don't care whether these answers are correct or not. The harm to the ecology of Stack Overflow forum is obvious.

However, it is not easy to identify such answers, and how to ensure that they should be sealed is a big question.

Many professionals can see the mistakes in this "false answer", but all they can do is step on it, and there is no guarantee that the content will be cleaned up.

Banning such content is a good thing, but what can the reviewer do? The above post said that the answers generated by these AI may look like high-quality answers, which means that for trained people, they may find mistakes and click on them.

The answers of these AI impersonators may have some likes and some steps, because they look like high-quality answers.

I don't know how to tell if an answer was generated by ChatGPT, and it now takes months for custom tags to be resolved, and these users may continue to damage Stack Overflow before they are processed in time.

So how rampant is ChatGPT?

Some netizens tried to post the question in the ChatGPT dialog box and compared it with the answer given by a "real person" to see if the generated answer could be used, only to find that something was obviously wrong with the "real person".

This is his own answer generated with ChatGPT:

This is the answer given by the so-called "human users":

It is not difficult to see that the two answers are the same in routines, the language tone is also very similar, and the code examples are almost the same.

This is clearly also a false answer generated by ChatGPT. At present, this "user" has been blocked by Stack Overflow.

In fact, from the point of view of OpenAI, the publisher of ChatGPT, one of the important uses of this AI tool is to help programmers check the code, and as a coding aid, answering questions on Stack Overflow is originally one of the legitimate uses of ChatGPT.

Unexpectedly, less than a week later, he was mercilessly banned. I don't know if I blame myself for being "not good at learning", the answer given is not helpful, or the netizens are so cunning that they are overburdened and broken.

OpenAI has been contacted by the media to ask how it plans to deal with the accuracy of these automatically generated answers, but there has been no reply yet.

Stack Overflow cannot be replaced yet.

As a developer, you must know the Stack Overflow that allows hundreds of millions of companies' programs to run and hundreds of millions of computer students to get their diplomas.

Here, you can find best practices for implementing certain functions, or solutions to obscure errors that occur only at 4 a.m. on a full moon.

However, after it was bought by Prosus for $1.8 billion in 2021, there was widespread concern about whether to open up the paid VIP model as well. And that day will probably be the end of programmers.

Fortunately, so far, the user experience of Stack Overflow is still excellent.

As for OpenAI's ChatGPT, it can not only answer questions in any field, but also give very accurate and detailed answers each time.

For example, when you ask "How to iterate over an keys and values array in Javascript", this is the answer:

To put the icing on the cake, ChatGPT can also tell you the best solution to this common problem in JavaScript through a supporting example and explanation:

So the question is, since ChatGPT can give an answer in a few seconds of typing a question, and it also provides code that can be copied directly, why don't we use it systematically?

Because ChatGPT has three "fatal" flaws:

1. There can be no mistakes in input

When using ChatGPT, you have to be precise, and even if you change a pronoun (a/an), the answer will be completely different.

For example, the previous question "How to iterate over an keys and values array in Javascript", this time we removed a "an" and became "How to iterate over keys and values array in Javascript".

Now the solution given by ChatGPT is only valid for simple arrays made up of a unified list of elements.

two。 Explanation is not humanized.

For example, the next question with the most votes on Stack Overflow:

"Why is it faster to operate on sorted arrays than on unsorted arrays? "

The answer given by ChatGPT is undoubtedly correct, but that's it.

Because, if you can understand this answer, of course you don't need to ask this question.

And if you need to ask this question, it means you can't quite understand the answer given by ChatGPT.

In contrast, interviewees on Stack Overflow will first present a general background in non-technical terms, and then gradually pull the topic back to the original and final questions to deepen thinking.

Of course, not all Stack OverFlow answers are (and need not be) so qualitative and detailed.

But this example fully illustrates that the big difference between a person and an AI is that the latter cannot determine the level of understanding of the other person and thus adjust the answer.

To put it simply, AI will explain the theory of relativity in the same way and in the same terms, whether you are a physics professor, an ordinary college student, or a rookie who knows nothing about it.

This is not important to AI.

3. The role of the community

Whether you like it or not, the biggest advantage of Stack OverFlow is its community. The creation and operation of this platform is to encourage and encourage the contribution of the overwhelming majority of people.

It is this wide variety of answers and opinions that allow users to determine their location and choose by considering the strengths and limitations of each solution.

In addition, another advantage of Stack OverFlow is peer validation. A solution that has been proven and tested by thousands of developers can provide a great guarantee for its correctness. (of course, this does not mean that 100% is correct.)

P.S. It can still be used.

There is no doubt that Stack OverFlow will never be replaced. It and GitHub will always be a good place to solve problems that require personalized explanation or difficult problems.

Like Copilot, Intellisense, and advanced IDE, ChatGPT (even if it's not limited to code) is just a tool in the developer's increasingly rich tool palette. Like any self-study model, it will learn and self-correct and improve over time.

But next time, instead of waiting a few days on Stack OverFlow for a relatively simple question (how to trigger the reconstruction of HookConsumerWidget in Flutter), ask ChatGPT directly.

After all, you have a good chance of getting a starting point that can be used to solve the problem:

Why did ChatGPT make so many low-level mistakes?

These days, all netizens have gone crazy with ChatGPT. Many people realize that this new AI is really a "force weapon"-it can generate a powerful and glittering answer for you without any effort.

If you put aside this gorgeous coat and look into it carefully, you will find that its answers are often full of mistakes and mistakes.

For example, I was 37 in 2022, so how old was I in 1985?

Because the age cannot be negative, there is no solution to this problem. ChatGPT:1985-2022 has no solution.

Well, is it possible that the author was born in 1985?

For example, this netizen asked ChatGPT, does Hobbes advocate the separation of powers?

ChatGPT gave the answer confidently and did not forget to quote the classics.

However, its answer is wrong.

Hobbes, a supporter of despotism, believes that the only viable alternative to anarchy is to give power to the monarch.

The idea of power distribution and checks and balances between the executive branch and the legislative branch was put forward by John Locke, a contemporary philosopher of Hobbes.

So why did ChatGPT make such a low-level mistake?

Obviously, this is because Hobbes and Locke are almost always mentioned together in political philosophy. When ChatGPT went online to find Hobbes' information, he naturally found Locke's explanation of the separation of powers, and then mistakenly pinned this view on Hobbes' head.

From this example, we can also see the limitations of the GPT-3 language model that underpins ChatGPT-it can only crawl, but it can't reason or think.

So, it's actually a matter of probability.

Recently, many people have been discussing: which jobs will be subverted by AI in the first place?

This is an open question. But at least we can get a hint of inspiration from the homework assigned by the teacher.

When the teacher assigns a political philosophy paper to the students, the student's achievement is usually just a reflection on what has been written a million times in the world.

Interestingly, AIGC's text is not like a math problem, it does not have the only way to the correct answer.

Yes, the output of AI is probabilistic: ChatGPT does not have any internal records marked right or wrong, it is actually a statistical model, which is a combination of languages in different contexts.

The context is based on the overall data set used for GPT-3 training, additional context from ChatGPT's RLHF training, prompt and previous conversations, and feedback soon.

The results of these combinations are undoubtedly exciting.

ChatGPT runs the entire virtual machine and writes code.

Of course, ChatGPT doesn't run python, but the answer is actually the probability results collected from the Internet data corpus that makes up GPT-3--

ChatGPT made the best guess of the result in 10 seconds, which was probably correct and felt as if a real computer was executing the relevant code.

This possibility is enough to surprise mankind.

Although it has a lot of gray areas for code work that requires accuracy.

But for other areas of AIGC, such as generating text and images, this multiple possibilities undoubtedly help to expand the boundaries of human imagination.

ChatGPT's business model to promote the revolution of artificial intelligence

Why is ChatGPT so popular these days?

GPT-3, as the underlying model, has been born for two years. In contrast, ChatGPT is free and easy to use.

It's one thing to read examples of AI output, and quite another to generate your own output.

The revolution of AIGC has already started. When Midjourney makes the art generated by AI simple and free, there is an explosive growth of human interest and awareness.

ChatGPT is free, which is critical.

You know, on OpenAI's API, the biggest constraint is cost.

Using OpenAI's most powerful language model, Davinci, to generate 750 words costs 2 cents; fine-tuning the model using RLHF or other methods can also cost a lot of money, 12 cents to generate 750 words from the fine-tuning model.

The charm of ChatGPT is that it makes OpenAI the same consumer AI product leader as MidJourney.

The business model of Midjourney subscription makes sense for things that have a marginal cost in terms of GPU time.

In this regard, the former GitHub's CEO Nat Friedman once put forward such an interesting point: in the real world, AI applications are actually scarce.

Now, researchers are providing a large number of new features to the world in a rapid way every day, and enterprises and product people are just beginning to digest these new features.

One step further, when AI is no longer scarce in the real world, will human beings be replaced by AI?

Let's take a look at AI's answer to this question.