"GPT-4.5" is here! The most powerful Code Interpreter on the surface is linked with Midjourney, 5 points to make a blockbuster

Source: Xinzhiyuan

Author: Taozi, so sleepy

Some time ago, GPT-4 became stupid and ignited public opinion. Some people even used up 25 quotas in 3 hours without solving the code problem.

However, after Code Interpreter (code interpreter) was officially lifted yesterday, netizens felt that ChatGPT is working again!

Even without writing/running code, it is much better than GPT-4.

Calling it a "plugin" is an understatement, it should be called GPT-4.5.

For the full launch of the "code interpreter", many Plus users are like digging treasures, showing the potential of the latest ChatGPT plugin they discovered.

Linkage with Midjourney, 5 points for a blockbuster

No, the "code interpreter" has begun to link with Midjourney, and the movie texture of the blockbuster is shocking.

A netizen said that he used a high-definition image generated by Midjourney to let ChatGPT generate the following video in less than 5 minutes.

Initially, he collected some hints from the web, but it didn't work for ChatGPT, thus allowing it to interpret the error.

He then asked if the "code interpreter" could create a zoomed-in video from the Midjourney image. It gives a step-by-step explanation of the process, along with detailed solutions.

All in all, you don't need to know how to create videos or anything. The Code Interpreter will ask you questions to clarify or help create ideas for the actions you specify. Let's look at other people's attempts.

Using the latest Panning function of Midjourney, the image can be extended from left to right multiple times. The panorama is as follows:

Then use ChatGPT to turn it into a video.

Netizen @huangyun_122 generated a video of his cousin talking about work late at night, and said that ChatGPT is still unprofessional in portrait and visual processing. Similar to the frame rate, the frame step size is not well understood, and the production is still not smooth enough.

：

The Q version of the penguin is so cute!

Most of the above netizens' ideas for creating videos come from the tutorials of netizen Chase Lean.

Yesterday, he used ChatGPT to generate a hamburger video on social media, which attracted the attention of many people. Even OpenAI co-founder Greg Brockman also forwarded this use case.

Immediately afterwards, he did another experiment, also turning the pictures generated by Midjourney into videos.

What's different this time is that Thanos finally dissipated, as if special effects were added.

To that end, Lean made his own publicly available on GitHub.

：

Star Wars Mini Game

In addition to changing pictures into videos, "code interpreter" can also make games.

Netizen Kris Kashtanova used ChatGPT to complete the following game within 5 minutes.

Co-founder Greg Brockman just retweeted this example of game development as well.

So, how is it done?

1. Input prompt: "Writing p5.js code for Asteroids, you control a spaceship with the mouse and shoot asteroids with the left mouse button. If your spaceship collides with an asteroid, you lose. If you shoot down all asteroid, you win! I want to make spaceships and asteroids with my own textures."

2. Go to the online editor Openprocessing to create, and save the sketch (you need to save it before uploading any texture files). Copy paste the code for GPT-4.

3. Generate a texture file, and remove the background, for example in Clip Drop.

4. Replace filename with your filename.

5. Run the program

6. If there is any problem, you can copy an error and paste it into GPT-4, let GPT-4 solve it.

7. To learn a little programming, write these prompts to GPT-4: "As my programming teacher. Tell me a detailed algorithm for an asteroid game, give the functions a name, and explain what each function will do. Don't Write the code. Can you then fully describe the algorithm for a 10-year-old?"

In short, netizens concluded that you can use any AI generator to make a game, and then let GPT-4 with a code interpreter write the code. If there are any issues, you can ask it to fix bugs.

Data Tabulator

In the past, to do a data analysis, a person needs various tools to draw the whole.

Now, with the "code interpreter", data analysis and icon production are all a matter of lip service.

To make a geographical icon of American lighthouses, you only need to upload the geographical location data, and then you can get the following visual GIF.

It can also effortlessly display public data, such as gathering data from public databases such as the International Monetary Fund, to make an intuitive icon.

For the data you want to visualize, after uploading the file, you only need to say "can you run some basic visualization", and a large wave of graphs will come.

"Code Interpreter" uses 10,000 lines of data to generate simple graphs

Creating radar charts has also become very simple. Netizens use the ChatGPT code interpreter to analyze a 300-hour Spotify favorite playlist, and get the following chart:

Also used ChatGPT to turn a CSV file of San Francisco crime data into a heatmap.

Emoji generator

Of course, ChatGPT can't escape generating emoticons.

The "code interpreter" can use existing Python libraries to draw visual effects, and then use GPT-4 to generate existing emoticons.

You can also tell it directly, "Create an entirely new meme by creating an image using the drawing tools available. The content is relevant to your experience working with humans. Give an example of a meme and explain how to use it and why very funny".

ChatGPT’s operation was as fierce as a tiger, and it directed and acted an emoticon package.

Someone else made it generate the first AI meme.

Prompt "Can you draw your own original meme and animate it? Make it fun and what it means to be the first AI meme".

"Code Interpreter" is GPT-4.5?

The weakening of GPT-4 has been hyped by netizens, and some people even call it together with GPT-3.5.

Regarding the fact that the "code interpreter" is called a plug-in, which does not match its capabilities at all, netizens compared the capabilities of GPT-4 and the "code interpreter".

Try running Donald Knuth's famous problem set on GPT4 and "code interpreter" and see the difference?

For another example, let them answer respectively: When and where the sun shines directly on Japan on July 4th.

GPT-4 analyzed for a long time, but did not give an answer.

The "code interpreter" finally gave an estimated time.

In addition, on the generated codes of the two, the specific comparison diagram is as follows.

Let’s talk about it together, have you discovered the potential of the “code interpreter”?