Organize | Tu Min
Listing | CSDN (ID: CSDNnews)
Is it worth $20 to hire a personal AI assistant that's really good at data analysis?
Now, OpenAI is making $20 more worth it.
In March of this year, OpenAI first announced the provision of third-party software application plug-ins for ChatGPT, allowing users to further expand the functionality of ChatGPT. And just last week, OpenAI announced in advance that one of the internal plug-ins, the ChatGPT Code Interpreter (Code Interpreter), will be released to users who pay $20 per month to open ChatGPT Plus.
Now, the long-awaited, everyone can finally use it.
The much-anticipated code interpreter is here!
The so-called code interpreter, that is, people who can't code, can only ask questions in natural language when facing almost all problems, and ChatGPT can also complete tasks that require complex programming skills. In reality, the ChatGPT code interpreter is not an ordinary AI plug-in.
According to OpenAI's description, the code interpreter allows ChatGPT to run code and optionally access user-uploaded files. You can also directly ask ChatGPT to analyze data, create charts, edit files, perform data operations, and more.
In addition, developers have also found that it is also possible to use the code interpreter to generate data visualizations and graphs, analyze music playlists, create interactive HTML files, clean data sets, and extract color palettes from images.
This code interpreter unleashes a myriad of capabilities that make it a powerful tool for data visualization, analysis, and manipulation.
At the same time, ChatGPT can also write code in Python and operate files up to 100MB.
This is just as netizens said, "Bring ChatGPT, I have the keyboard, and go anywhere in the world."
And as Wharton professor Ethan Mollick commented, "Using the new ChatGPT tool, things that took me weeks to master during my Ph.D. can now be done in seconds, and the mistakes are usually faster than human analysts." less".
Even Linas Beliūnas, the European country manager and general manager of Lithuania of the well-known financial company Flutterwave, also commented on his LinkedIn, "OpenAI is unlocking the most powerful functions since GPT-4 to everyone. Now anyone can become a data analyst. .”
Now, to try it out, you first need to have a ChatGPT account, then pay $20 to buy ChatGPT Plus, and then activate the code interpreter and other experimental new features in the account settings, and you can "get on the bus".
So, is this code interpreter really that magical? Based on this, users at home and abroad have summarized multi-dimensional getting started guides and various usage postures on the basis of continuous attempts. Let's take a look together.
Code Interpreter + Midjourney: “Great Death Four Directions”
Many netizens tried it directly with Code Interpreter, and the result successfully made an image quickly become a film within 30 seconds!
To ask how it was done, @Chase Lean shared his "five-step" experience of using a code interpreter on Twitter:
One is to open the Code Interpreter function in ChatGPT;
The second is to upload the image you want to animate, such as:
The third is to tell ChatGPT your needs, such as: animate this food image from left to right;
The fourth is to give ChatGPT some clear tips, such as modifying the size of the picture, and the specific display rules:
Fifth, press the Enter key, and after waiting for 30 seconds, ChatGPT will automatically provide the download link of the video.
In the end, the following effect is obtained, isn't it amazing:
If only one image is needed to generate a video, can the image itself also be done with the help of AI tools? What kind of sparks will the powerful combination of AI and AI tools create?
Netizen @黄谣 (https://twitter.com/huangyun_122) combined the text-to-image tool Midjourney with the code interpreter, and finally came to such a conclusion: "It's amazing, one-time production! My only worry What I dare not face has finally come."
He got such a video effect by asking MidJourney to generate a 36:9 wide-format image, upload it to Code Interpreter, and ask it to be displayed in the order of "left, center, right, center, and left":
In this regard, he commented:
Code interpreters can also be "beautiful on their own"
Of course, in addition to effectively combining some mainstream tools in the industry, the code interpreter itself also has a lot of functions.
When it comes to specific use, data analysis is the most commonly used function of code interpreters. Its proven ability to quickly analyze large amounts of data and provide detailed insights makes it an invaluable tool for professional data analysts and amateurs alike.
Professor Ethan Mollick from the Wharton School of Business wrote a detailed guide, and took the code interpreter to process data as an example, and shared the whole process of getting started and experience in depth.
He used a superhero dataset (https://www.kaggle.com/datasets/claudiodavi/superhero-set) on Kaggle, a data modeling and data analysis competition platform, as a reference, and downloaded this dataset.
Then click the "+" button on ChatGPT to upload data, or directly upload compressed data such as ZIP. Then give an appropriate initial hint, and the AI can figure out the meaning and structure of the data from the context.
Ethan Mollick found in the process of using the code interpreter that this tool is more of a conversation with AI, rather than going all the way down through the form of Prompt.
In the link of using the code interpreter, Ethan Mollick said: Prompt seems to only play an important role in two dimensions:
One is that AI sometimes forgets that it can achieve some functions, such as making GIF or 3D drawing. At this time, you can "encourage" it through Prompt (such as: "You can make GIF, please try").
The second is that when you want AI to improve its work, you only need to provide some simple prompts such as "further test this result" or "make this chart better".
When the data is successfully uploaded, the display is as follows:
Next, let ChatGPT do the tedious part of any data analysis job: data merging and cleaning . Human users only need to let the AI get started directly through short prompt words.
In this session, Professor Ethan Mollick said that often times, AI will correct its mistakes when it finds them. But to be on the safe side, it is recommended that human engineers also participate in the verification instead of blindly trusting artificial intelligence.
Now, on to the analysis part.
AI seems to know a lot about analytics, Ethan Mollick found. He expressed interest in doing some predictive modeling, which would predict what abilities a hero might have based on other factors. And ask how ChatGPT should handle this problem?
Unexpectedly, ChatGPT directly built a random forest classification model, which is very efficient.
At the same time, Ethan Mollick expressed his unwillingness to allow ChatGPT to estimate missing data by using numerical data. When further asking AI whether there is a better method or alternative, ChatGPT can also provide suggestions based on its extensive knowledge.
Ethan Mollick concluded that the most impressive aspect of Code Interpreter is that it "reasons" on data in a very human-like way. When asked about the results of the network analysis, ChatGPT came to an interesting conclusion: the set of abilities that heroes typically have is visual in nature (as they come from comic books), fits certain archetypes, and is best for building ongoing adventures . A neat way to integrate data and stories together!
5 minutes to make a game, create a diagram... code interpreter is not a problem
At the same time, the code interpreter can also make games, create diagrams, convert any GIF to MP4 with one click, and more.
@Kris Kashtanova shared on Twitter how he spent 5 minutes making a game with the help of the code interpreter feature in ChatGPT.
details as follows:
1. Write a Prompt like this: "Write p5.js code for Asteroids, control your spaceship with your mouse, then left click to shoot an asteroid. If your spaceship collides with an asteroid, you lose. If You shot down all the asteroids, you win! I want to use my own textures for the spaceship and the asteroids."
2. Go to the Openprocessing website to create and save the sketch (you will need to save it before uploading any texture files). Copy-paste code from 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 something goes wrong, GPT-4 can fix it (you can just copy the error and paste it into GPT-4), just like asking a human programmer.
7. To learn some programming, write these prompts into the GPT-4 dialog window: "Be my programming teacher. Tell me the algorithm of the Asteroids game in detail, naming the functions and explaining what each function does. Don't write code yet. Then Can you describe this algorithm to a 10-year-old?"
In the end, a small game was easily born like this:
However, it should be noted that the arrival of this most powerful code interpreter can save you from writing code, but it does not mean that you can stop learning coding. For some difficult program generation and design, such as the above-mentioned game development, it is still necessary to have a certain game development and programming background in order to correctly debug and realize the final program operation.
Compared with this difficult time, some netizens have summarized some practical examples based on the ChatGPT code interpreter:
Create various charts
As demonstrated by OpenAI co-founder Greg Brockman, just ask gently, "Can you draw 10,000 random points on a plane? Each dimension is normally distributed", and ChatGPT gives such a result :
Convert any GIF to MP4 in one click
Upload a GIF image, ChatGPT can generate an MP4 video file back!
create palette
Also based on ChatGPT, it is also possible to extract colors from images to create simple PNG palettes.
One-click data visualization
According to the cleaned data, the content of the data visualization chart is directly generated:
Convert CSV file to GIF
A user has uploaded a CSV file with the location of every lighthouse in the United States. ChatGPT directly creates a gif of the map of lighthouse locations after a few seconds, where the map is very dark, but each lighthouse is blinking.
Analyze playlists to generate unique music tastes
Another user analyzed 300 hours of Spotify favorites playlists using the ChatGPT code interpreter. Finally, GPT showed him how to export from the Spotify API, run multidimensional PCA and t-SNE analysis, and summed up his taste for him:
Create fully functional HTML heatmaps from datasets
Have you used this code interpreter yet?
Based on the above, this is just the surface of the code interpreter practice, and more functions need to be unlocked continuously. But just through the above few examples, we have witnessed that AI is freeing more practitioners from the most annoying and repetitive parts of their work, allowing them to focus on more valuable things.
In this regard, Ethan Mollick, a professor at the Wharton School of Business, said, "My time has become more valuable, not less, because I can focus on important things instead of rote memorization. The code interpreter represents so far. AI's clearest, most positive vision for work: Disruption."
Clearly, Code Interpreter is clearly setting a new standard for the future of AI and data science. And with this tool, OpenAI has once again pushed the boundaries of ChatGPT and Large Language Models (LLMs).
reference:
https://twitter.com/icreatelife/status/1678184683702566922
https://twitter.com/chaseleantj/status/1677651054551523329
https://venturebeat.com/ai/code-interpreter-comes-to-all-chatgpt-plus-users-anyone-can-be-a-data-analyst-now/
https://medium.com/@smraiyyan/chatgpt-code-interpreter-launches-next-week-here-are-10-mind-blowing-use-cases-c20ce9be3824
