ChatGPT steals: Stack Overflow is being abandoned by programmers, and the number of visits plummets by 32 million in a month

This article is reprinted with the authorization of AI New Media Qubits (public account ID: QbitAI). Please contact the source for reprinting.

Stack Overflow is being abandoned by programmers.

You heard it right.

The number of visits to this world-renowned developer Q&A website dropped by 32 million in just one month!

Even the current search volume is only one-third of its peak...

Why is this suddenly happening? A netizen explained the mystery in one sentence:

I haven’t used Stack Overflow since ChatGPT came out.

And judging from the Stack Overflow traffic statistics in the past three months, the traffic decline occurred between November and December 2022.

Coincidentally, OpenAI released ChatGPT on November 30 last year.

Good guy, it turns out that it’s ChatGPT, the most popular hot potato in the AI ​​​​circle.

Then why can it change the knowledge-seeking habits of many programmers so quickly in such a short period of time?

Regarding this issue, programmers, as users, may have the most say.

Someone summed up the problem concisely and concisely into two points:

If I am not lazy: Google and Stack Overflow are the first choices;

If I am lazy: I choose ChatGPT.

#The implication is obvious. The key to the problem lies in the convenience of getting the answer.

Let’s first look at the process of getting answers on Stack Overflow.

First of all, you need to click the "Ask Question" button to start the questioning journey:

Then you come to the tedious "fill in the blanks" link:

Enter the title → Describe the problem → What have been tried/what results are expected → Language tag

But this series of operations is not over yet, for the questioner , the most helpless thing may be the long wait.

However, the reality is often like this:

That’s right, no one cares about it; and even if someone answers the question, you still have to look at it Which answer has the highest "vote" and so on.

But if this matter were done with ChatGPT, the result would be different. It only requires one action:

Ask!

Then the answer will be thrown in front of you with a "snap".

After such comparison, the convenience of ChatGPT in "question and answer" can be seen. As netizens summarized:

Search for products/websites→ Google

Seeking answers→ ChatGPT

But there is one thing to say, "Q&A "Convenience is certainly one aspect of this matter, but the more important point should return to the accuracy of the answer.

Then the next question is:

Is it reliable for programmers to rely on ChatGPT?

You must know that the rationale given by Stack Overflow for banning ChatGPT was:

The purpose of (doing this) is to slow down the large number of answers created using ChatGPT from flowing into the community.

Because the probability of getting wrong answers from ChatGPT is too high!

However, this is only a one-sided statement from "the opponent" and cannot be fully believed. It's been a while since ChatGPT was released. Whether it's reliable or not, let's take a look at how users feel about it.

Let’s talk about the conclusion first. ChatGPT can be said to be an “all-round talent” in terms of programming: finding bugs, writing websites, converting prompt words into codes...

ChatGPT can be said to be quite considerate in finding bugs. For example, the CEO of Repit dumped a piece of erroneous code to ChatGPT and asked it to find the bugs in it.

As a result, ChatGPT not only found the bug, but also pointed out the cause of the error, how to modify it, and attached the correct code. Finally, it did not forget to summarize.

(Isn’t this better than searching on Stack Overflow to find answers? Manual dog head)

Of course, an AI model cannot compare with Stack Overflow, a question and answer website. , let’s see how it compares with other code repair systems.

Not long ago, Johannes Gutenberg University and the University of London also conducted a study in which ChatGPT compared three other code repair systems: Codex, CoCoNut and Standard APR.

During the research process, the researchers gave a total of 40 error codes, and the preliminary results obtained are:

ChatGPT solved 19 Questions, Codex solved 21, CoCoNut solved 19, and Standard APR method solved 7.

Among them, Codex and ChatGPT come from the same language model family, so the results are relatively close.

BUT! This is not the final result. After all, ChatGPT is a conversation model and can communicate. After the communication, it solved as many as 31 bugs.

Well, the final result is also obvious.

In addition to debugging, ChatGPT can also automatically write code when given requirements. Netizens have reported that the final code effect is not bad.

For example, some netizens claimed that most of the code of their startup company was completed by ChatGPT and Copilot.

There is also a "student fan" of ChatGPT who wrote a website using ChatGPT. The generated code is very neat:

Even at the end of last year, Riley Goodside received an offer from Scale AI, a Silicon Valley unicorn company valued at US$7.3 billion, by playing with ChatGPT prompt words, and officially hired him as a "prompt engineer."

The content of the work is: just speak a few words and let AI generate the code you want.

There are too many examples to list here. In short, some people have now incorporated ChatGPT into programming productivity.

Moreover, ChatGPT is also a good model for “improving”. It was previously rumored that it had insufficient mathematical capabilities. Not only did OpenAI officially announce yesterday that it has improved ChatGPT’s performance in terms of authenticity and mathematical capabilities.

For programmers, what they are more concerned about is whether ChatGPT has improved in programming capabilities.

However, in this regard, some netizens have tested it. The same code problem could be answered at the end of December last year (on the left), but not at the end of January this year (on the right).

One More Thing

Recently, ChatGPT has "new equipment". It has just released a model BLIP-2, which can be connected to ChatGPT. It will not only You can simply look at the pictures and speak, explain the plot, and add words to the pictures...

Moreover, BLIP-2 has achieved SOTA on various visual language tasks, and the code is now open source.

Joining forces, looking forward to a wave~

Reference link:

[1] https://twitter.com/dannypostmaa/status/1620207540381569024

[2] https://techcabal.com/2023/01/31/stack-overflow-chat-gpt/

[3] https://twitter.com/flaviocopes/status/1620333315919331328

[4]​ https://twitter.com/laminappcom/status/1620516951163559936

[ 5] https://www.pcmag.com/news/watch-out-software-engineers-chatgpt-is-now-finding-fixing-bugs-in-code

[6] https: //twitter.com/mrdbourke/status/1620353263651688448

The above is the detailed content of ChatGPT steals: Stack Overflow is being abandoned by programmers, and the number of visits plummets by 32 million in a month. For more information, please follow other related articles on the PHP Chinese website!