Today, the ChatGPT "code interpreter" is officially lifted! Turn pictures into videos in 30 seconds, and make watches with your mouth | The Complete Works of Ten Amazing Magic

Source: Xinzhiyuan

Waiting for more than 3 months, the ChatGPT "code interpreter" is finally fully open, and a wave of magic is coming.

Families, it's finally here!

Just now, the ChatGPT "code interpreter" beta version is officially open to all Plus users.

Karpathy, a scientist at OpenAI, also publicized this, and the beta version of the code interpreter is quite powerful.

It is your personal data analyst: can read uploaded files, execute code, generate charts, statistical analysis and more. I expect it will take some time for the community to reveal its full potential.

Some time ago, the decline of GPT4's ability caused an uproar on the Internet, and some people even thought it was a little stronger than GPT3.5.

And now, the code interpreter makes people feel again, the familiar and powerful GPT-4 is back!

As soon as the plug-in was launched, many netizens started to test it.

Function unlock

So, what exactly can a "code interpreter" do?

30 seconds, the picture becomes a video

No, some netizens actually used ChatGPT to turn pictures into videos. This effect is even more explosive than Midjourney and Runway.

How is this done? Netizens gave a full set of tutorials.

Step 1: Open the "Code Interpreter Model". To enable it, go to Settings -> Beta Features -> Code Interpreter.

Step two:

Select the Code Interpreter option from the drop-down menu. Then, upload the image you want to animate.

Step three:

I want to show some food from a fancy restaurant. So, I asked ChatGPT to animate this food picture from left to right. By modifying my GPT hints (see below), you can create your own animations.

details as follows:

Step four:

Press enter. After 30 seconds, ChatGPT will give you a download link of the video.

Co-founder Greg Brockman retweets the code interpreter for image conversion when it's impossible to remember ffmpeg calls.

Facts have proved that the "code interpreter" is really good. Now, if you cut a video, it only needs one sentence to get it done.

"There is a certain memory capacity limit, but it is no problem to process small files. I just threw a 4-second gen-2 video to TA, intercepted the first two seconds, no problem, and it was done quickly.

It will automatically test the appropriate code. If you want to program, you can throw a file to ta and ask for it, and then look at the code provided by ta.”

### PDF to image

"Bao Yu", a well-known Internet technology blogger in China, used a "code interpreter" to convert PDFs into images.

In the past, it was necessary to combine the content of each page into one picture, but now, one can do it.

### GIF to video

In addition to converting pictures to videos, you can also convert GIFs to videos.

Netizens gave ChatGPT a hint: I upload a GIF, you give me a 5s MP4, and use a story lens to zoom in slowly.

Upload GIFs

Finally ChatGPT generated the video immediately

, duration 00:05

In addition, ChatGPT can also make a "matrix" GIF.

### Generate QR Code

Two days ago, OpenAI announced that the ChatGPT "code interpreter" will be fully open.

It shows an example that can generate a QR code. In just one sentence, you can complete the task by simply moving your mouth.

### "Pocket" Data Analyst

The biggest use case for code interpreters is data analysis.

For example, let it analyze the life expectancy data of various countries, the largest cities in the world, (you can upload CSV, word, PDF or video files)

Then, talk to the data and ask any questions you might want to ask.

Or ask to extract unique random trends from your data for a surprise.

For example, find unique trends in two data sets and display them in a tabular form.

The final step, visualization.

Turn data into any chart or graph you want. Best of all, it's just one click away.

For example, generate a scatterplot illustrating the relationship between a country's GDP per capita and its average life expectancy.

Seeing such a powerful "data analyst", netizens started to RIP again.

user's guidance

How to use "code interpreter"? A Wharton professor, Ethan Mollick, wrote a guide just for this.

First of all, Mollick pointed out that the name Code Interpreter continues OpenAI's long tradition of giving their own products bad names.

Because—although it is called a "code interpreter", in fact it is most useful to those who do not understand programming at all.

Essentially, the "code interpreter" allows GPT-4 to upload and download information, and write and execute programs for the user, which allows AI to do all kinds of things it couldn't do before.

Specifically, it provides AI with a general problem-solving toolbox (by writing code in Python), large memory (up to 100MB, can be in compressed form), and users can integrate this toolbox into AI to play LLM The advantages.

This helps to fix many issues with previous versions of ChatGPT. for example,

It allows AI to handle math (even if it's complex), but also to handle words more accurately (such as actually counting words in a paragraph), and the underlying reason is that it can write Python code to address LLM's natural weaknesses in mathematics and language.

For example, like the picture below.

As you can see, the "code interpreter" can accurately count the number of words in the paragraph.

In addition, the "code interpreter" reduces the chances of hallucinations and nonsense.

When an AI uses Python directly, the code helps keep it "honest". The reason is also very simple, if the code is incorrect, Python will report an error. Since code operates on data, AI cannot insert errors into the data.

Of course, it's not perfect. There is still a chance that the AI will hallucinate, but this is not common and is unlikely to affect the code or data itself.

In this way, AI is closer to general artificial intelligence - a large number of problems can be solved by code.

Moreover, GPT-4 is very good at deciding when to use a "code interpreter" in new and interesting ways.

Mollick gave an example, if he had GPT-4 prove to the user that the earth is round, then a "code interpreter" would be used, providing multiple arguments, combining text with code and images.

Previously, people who needed programming in various majors had to know how to write code, but they had to debug it by themselves after writing. Even if the previous AI can provide some help, it is still quite difficult for people who do not have a basic knowledge of Python, and they have to use the power of AI to modify the code repeatedly.

And now, with the "code interpreter", AI can correct errors and debug by itself!

Then, Mollick talked about how to use the "code interpreter" to process data, which can be regarded as a real pain point for a large number of users.

Mollick said that the "code interpreter" operates at a very advanced level, can automatically handle many complex quantitative analysis, and can adopt very complex data methods.

Users can easily upload data, even compressed data such as ZIP files, just by clicking the plus sign.

At the same time, when providing data, it is best if the user can give one. AI can then figure out the meaning and structure of the data from the context!

In addition, Mollick pointed out that when using the "code interpreter", the user is more in a "dialogue" with AI, and does not need to rack his brains to write a rigorous and concise one.

Just like chatting with a data analyst, how you talk to people is how you communicate with the "code interpreter".

But there are two special cases here, and you still need to write something similar.

One is that AI sometimes doesn't remember what it can do, and at this time it needs one to wake up the AI.

For example: "You can make GIFs, please try."

Another situation is that when the user is not satisfied with the result, the "code interpreter" can be used to optimize the result.

For example: "Please run further tests on this result", or, "Can this graph be made better".

After loading the data, GPT can perform data merging and cleaning operations. The specific processing method is quite complicated, but as mentioned above, just think of the "code interpreter" as a face-to-face data analyst.

At the same time, Mollick reminds users that it is best to check the results after the AI corrects itself and corrects itself, in case something goes wrong.

Immediately after Mollick practiced it, he asked the "code interpreter" to make a predictive model to predict what superpowers a superhero might have.

As you can see, the "code interpreter" builds a Random Forest Distributor.

Of course, in the actual operation process, users still have to "communicate more" with GPT. Mollick doesn't want AI to use numerical data to estimate the missing data. Mollick wants to discard the missing part directly.

But he pointed out that it is also possible to talk to GPT and let it change to a different processing method.

Afterwards, Mollick fed GPT a "Can you do another complex and interesting analysis?" question. He said he was impressed by some of the analysis methods chosen by Code Interpreter.

Just like the picture below.

"Code Interpreter" cleverly analyzes the network situation of superpowers. As you can see, the methods are diverse and very interesting.

The most impressive thing about "code interpreter" should be its reasoning ability. Mollick summed it up as very "human".

Plus, the level of interactivity in visualizations is still online and improving. Users can continuously let AI improve or change the output content.

Mollick had the AI create an interactive table, including six charts, including one in 3D.

In order to make the chart more beautiful, Mollick continued to input and adjust the results.

The final result looks like the one on the right in the image below.

Top 10 Amazing Magic

Netizen @GPTDAOCN recently summarized the top ten shocking use cases of ChatGPT:

Visualization of the code interpreter

1. Convert GIF to MP4: It can convert dynamic pictures (GIF) to video files (MP4), making it easier for you to share and use.

2. Create various graphics: You can use this function to generate various graphics such as histograms, line charts, pie charts, etc., to visually present and analyze your data.

3. Extract colors from image to create PNG palette: This function can extract the main colors from any image, and then generate a palette in PNG format.

4. Clean data to generate in-depth visualization: If you have some messy data, this feature can help you clean and organize it, and then convert it into a clear visualization.

5. Convert CSV file to GIF: This function can display the data in the CSV (comma-separated value) file in a dynamic way and generate a GIF image.

6. Analyze playlists to summarize your music taste: You can use this function to analyze your music playlists to summarize your music taste.

7. Convert your data set into a fully functional HTML website: This feature can quickly convert your data into a fully functional website for you to display and share.

8. Create a downloadable stock data set: This function can help you create a stock data set, and this data set can be downloaded for your in-depth analysis and research.

9. Convert images to text files: If you need to extract text from images, this function can help you convert the text in images to text files.

10. Detailed analysis of unformatted data: For those raw data without a specified format, this function can help you conduct in-depth analysis to obtain valuable information.

In fact, the use cases of code interpreters are much more than these.

Comments from netizens

I'm trying ChatGPT code interpreter. So far, all I can suggest is "every developer start working on it now".

How many people's lives have been cut off by Code Interpreter? Too many administrators process data inefficiently every day, and now only need natural language to automate the office.

References: