Home > Article > Technology peripherals > ChatGPT can run its own code: when you submit a request, you can directly input the running results. Netizens called it "magic" after the internal test.
Once something written by ChatGPT has a bug, you must immediately restart a new session, otherwise its errors will begin to multiply: this feels like it is trying to hide something, which is very strange.
If you give ChatGPT a request, it can only give you a bunch of code with unknown operating effects?
Now, with just a small change, ChatGPT can not only directly turn your text requirements into code, but can also run it for you and return the output results to you!
The secret is the code interpreter plug-in.
Writer Andrew Mayne (now joining OpenAI) obtained the qualification for internal testing and tried it. The effect made him call it "magic"——
Let ChatGPT write an AI face detection program. After Andrew Mayne uploaded his own photo, it was quickly detected:
Try again and let it write a maze generation algorithm and make the path into a Pac-Man animation. It’s no problem at all:
Some netizens saw this internal beta Sigh after the article:
This is so crazy, it will change everything.
Let’s take a look at what magical “magic” Andrew Mayne discovered in the closed beta.
As one of the most popular ChatGPT plug-ins, the code interpreter is a Python interpreter that works in a sandbox and firewall execution environment and contains some temporary disk space.
To put it simply, you only need to enter a sentence into ChatGPT. It can not only write the code, but also use the interpreter to run through the code, give an explanation of each line of code, and output the result to you:
The generated content is also diverse, from text, image and sound processing to chess game engines and simple AI algorithms, ChatGPT can handle it.
Let’s first look at the generation and processing effects of images.
For example, use ChatGPT to convert a Cthulhu image to ASCII:
Generate a cat wearing a hat and holding a pipe (a bit abstract):
Generate website QR code:
It’s no problem to make a relatively simple animation, such as "Blizzard":
Let’s listen to the sound again.
For example, to generate a Shepard tone (a pitch that sounds like it is rising or falling, but is actually just an infinite loop of pitches):
Not only gave an explanation, but also generated a tone, which sounds like it really means something:
Sheppard tone audio generated by ChatGPT: 00:0000:10
Finally, there is text processing. Take the OCR algorithm as an example. This is an algorithm that recognizes pictures into a piece of text:
After uploading an old photo, AI quickly recognizes it The corresponding text was output and converted into a .txt file, which looks pretty good:
So, go one step further and try using ChatGPT to write the algorithm?
In addition to the face recognition algorithm mentioned at the beginning, which was generated by ChatGPT based on OpenCV, Andrew Mayne also tried the NLP algorithm this time.
Specifically, half of the sentence is given and the AI predicts the next word. ChatGPT quickly gave the code:
It looks good and the generated effect is how?
Andrew Mayne tried a sentence:
They went to the... (They went to...)
Usually this should be followed by a location noun, such as a beach or a church, etc. wait. However, after thinking about the AI algorithm written by ChatGPT, it answered "shoulders"? ? ?
They went to the shoulders. (They went to the shoulders)
Andrew Mayne joked that there is no need to worry about ChatGPT writing a GPT-4.
The last step is to generate the game engine. Andrew Mayne said "it's absolutely no problem", but did not give specific steps for ChatGPT.
This is the effect of generating a checkers engine (in addition, it can also generate a chess engine):
It can also be used to generate a cellular automaton (Game of Life):
Some netizens said after reading it that the code interpreter plug-in fills in the ChatGPT very well" Bad Math” vulnerability:
ChatGPT has almost no math skills, but it can easily do it by generating Python code.
#But many netizens found many bugs during this demonstration.
For example, although they are all Python programs that can be run, there are basically small errors that are obvious if you look closely, including:
However, "you don't need to worry too much about these obvious bugs."
Some netizens feel that what really needs to be worried about is the ChatGPT "overlay" error" question.
Once something written by ChatGPT has a bug, you must immediately restart a new session, otherwise its errors will begin to multiply:
This feels like it is trying to hide something, very strange.
Finally, although ChatGPT can already write a chess engine, some sharp-eyed netizens discovered that it still cannot play chess.
For example, suddenly use a horse to eat your own soldiers. (Manual dog head)
Reference link:
[1]https://andrewmayneblog.wordpress.com/2023/03/23/chatgpt-code-interpreter- magic/
[2]https://news.ycombinator.com/item?id=35312609
[3]https://twitter.com/peteskomoroch/status/1639894112307679232
The above is the detailed content of ChatGPT can run its own code: when you submit a request, you can directly input the running results. Netizens called it "magic" after the internal test.. For more information, please follow other related articles on the PHP Chinese website!