🫱Click here to join the group chat of 18 subfields (🔥Highly Recommended)🫲
Since GPT-4 came out, my code has been different. The good news is that I can save 5 hours of work each week. The bad news is that I might completely forget how to program.
—— Data Analyst Ken Jee
Recently, the application of ChatGPT in the field of programming has attracted widespread attention. Many programmers have found that their coding methods have changed significantly after using ChatGPT, and these changes have undoubtedly brought positive effects. Advanced technologies like ChatGPT and GPT-4 are expected to fundamentally change our perception and approach to programming.
▲Figure 1｜Data Analyst Ken Jee using GPT to assist in programming ©️【Deep Blue AI】CompilationBy using ChatGPT, programmers have significantly improved their efficiency, and debugging time has been greatly reduced. Compared to the previous coding process, this change allows for saving about 5 hours each week. So, how are these efficiency gains achieved?
This article introduces three main methods to harness the power of ChatGPT to improve programming efficiency. I hope that those in need of programming can draw inspiration from these experiences and illustrated examples to further master and apply these skills, enhancing their programming efficiency and quality.
In modern data work, data analysts are gradually breaking free from the constraints of the cold start problem. Nowadays, they no longer need to write code completely from scratch but can cleverly use templates or existing template code as a starting point.
▲Figure 2｜GPT can write startup programs for projects ©️【Deep Blue AI】CompilationFor exploratory data analysis, for example, if we want to quickly plot histograms, box plots for each variable, create correlation matrices, and check the number of null values in each column of the dataset, ChatGPT can easily generate template code for these tasks, and we just need to fine-tune it according to the specific dataset.
Imagine, a piece of code that might originally take 20 minutes to write manually can now be completed in just 5 minutes with the help of ChatGPT.
The cold start problem often leaves us lacking motivation when facing daunting tasks. That feeling of “everything is hard at the beginning” is like a marathon runner standing at the starting line, feeling anxious and hesitant about the long 26.2-mile course ahead. Similarly, sometimes looking at a blank notebook, we may feel somewhat at a loss, knowing that there is a lot of work waiting for us before the project takes shape.
However, with tools like ChatGPT, we can more easily take that crucial first step, making data analysis smoother and more efficient.
▲Figure 3｜Marathon runners at the starting line ©️【Deep Blue AI】CompilationTo solve this problem, past experiences suggest that an effective method is to copy and modify some existing code from other projects as a starting point.
However, to be honest, nowadays everyone prefers to use ChatGPT to solve this problem. The way ChatGPT outputs code directly onto the screen makes engineers feel like they are writing the code themselves, which greatly enhances the sense of participation and accomplishment, as if our contributions far exceed the actual workload.
This feeling indeed injects a lot of motivation into work. What’s even more exciting is that even if ChatGPT cannot continue in some places, we can easily restart from the breakpoint and quickly generate various changes and adjustments.
Have you ever had the experience of having a clear idea in mind but struggling to turn it into code?
▲Figure 4｜GPT-4 can help realize programming ideas ©️【Deep Blue AI】CompilationThe emergence of ChatGPT, especially GPT-4, has greatly solved this problem. It has made significant improvements in understanding and translating casually input ideas into code compared to previous versions.
This method, which we call “conversational programming,” is gradually changing the traditional mode of programming.
In the past, you might need to search the internet for suitable code snippets based on an idea and then adjust and debug them according to your needs.
Now, you only need to simply describe the functionality you want, and ChatGPT will generate code that closely meets your needs. If there are issues with the code, you can directly tell it, and it will quickly make modifications.
▲Figure 5｜GPT-4 writes code for a machine learning project, predicting the winners of the PGA Tour ©️【Deep Blue AI】CompilationFor example, to create a simple function that adds two numbers, while this sounds simple, let’s assume you are a beginner.
ChatGPT can quickly generate this code and, upon your request, modify the function to handle lists of numbers or the sum of two lists. This interactivity is impressive; it can learn and optimize based on your feedback.
▲Figure 6｜GPT-4 helps quickly generate functions ©️【Deep Blue AI】CompilationMoreover, the conversational programming approach with ChatGPT also helps people use their time more efficiently.
Users’ thinking speed often exceeds their coding speed, and real-time conversations with ChatGPT allow us to focus more on thinking without frequently switching between thinking about problems and writing code.
This way, users’ critical thinking skills are better exercised, while the speed of turning ideas into code is greatly enhanced.
Let’s compare these two processes:
▲Figure 7｜Comparison of programming methods before and after using GPT ©️【Deep Blue AI】CompilationIn the past, programmers often had to repeatedly think through the code writing process. However, now, through communication with ChatGPT, they can form thoughts, execute, and rethink in an iterative loop more quickly, significantly shortening the time required for this process.
▲Figure 8｜Programmers using GPT save work time ©️【Deep Blue AI】CompilationFrom the above image, it can be seen that with the assistance of ChatGPT, programmers have made significant progress in the same amount of time and have reduced a lot of blanks in their thought processes, making the programming process more efficient and smooth.
When it comes to functions, they hold a pivotal position in programming. The essence of this concept lies in writing efficient and high-quality code once for reuse when needed, thus avoiding redundancy and repetitive work.
In the daily practice of data analysis, while the use of functions is quite common, during certain stages, such as testing, it may be more convenient to run code directly. Especially when facing a lot of iterations, encapsulating code into functions may add extra debugging work, as it means debugging not only the code itself but also the use of the function.
However, we have recently found that the use of ChatGPT is quite similar to the concept of functions. Taking the training of an SVM regression model as an example, we do not need to rewrite the same code for each model or manually replace variable names; we can use ChatGPT to accomplish this task.
Specifically, the user only needs to provide a basic code snippet to ChatGPT and request it to generate corresponding style codes for decision trees, random forests, and XGBoost. Surprisingly, ChatGPT not only completed the task quickly but also accurately replaced all variable names, exceeding expectations.
▲Figure 9｜GPT-4 programming implements decision trees ©️【Deep Blue AI】CompilationFurthermore, ChatGPT can even integrate these codes into a function that accepts models as parameters and outputs relevant model metrics. This feature greatly improves programming efficiency and convenience.
Although conversational coding has brought significant changes to data science workflows, this functional application method is undoubtedly one of the most time-saving and labor-saving methods to date.
Before the release of GPT-4, many improvement attempts seemed exceptionally difficult and sometimes even impossible. At that time, conversational coding and code functionalization seemed rather clumsy. The time required to debug code generated by ChatGPT was almost equivalent to completing all work without it.
GPT-4 has shown significant improvements in model performance for coding use cases. However, no technology can achieve perfection.
The most significant advancement of GPT-4 is its ability to have a certain level of self-awareness, recognizing the users it interacts with. However, the accuracy of user data still needs improvement. The model may still produce misleading information in some cases, with about 20% of instances potentially providing incorrect information.
▲Figure 10｜GPT-4 can recognize interactive users ©️【Deep Blue AI】CompilationTherefore, when using the code generated by GPT-4, one should remain vigilant and carefully check its accuracy.
While the code generation process has become more convenient, it is also important to recognize that it may sacrifice some personal style and human creativity, which are the unique traits of human work.
Additionally, potential ethical issues and the propagation of biased algorithms are also worth noting. When using GPT-4, it is necessary to ensure that the generated code aligns with the corresponding values and principles.
【References】
https://medium.com/mobilepeople/how-well-can-you-use-chat-gpt-as-your-coding-assistant-fe6fed0e02bc
https://youtu.be/RRnThAtKuq0?si=S1N_voj_vxqq4gd5
https://towardsdatascience.com/3-great-ways-use-chatgpt-gpt-4-better-coding-7fb94e86be3e
https://arxiv.org/abs/2304.13187
Newcomer in the 3DGS world? Achieving the fastest training speed to date, over 6 times faster!
2024-06-09
HKU Open Source: Achieving automatic removal of dynamic occlusions in 3D reconstruction environments!
2024-06-07
【Invitation】
【Deep Blue AI】 opens an authorized sharing channel, inviting laboratories and individuals in the fields of 【AI + Autonomous Driving + Robotics】 to submit authorization, hoping to provide a more convenient platform for readers to communicate with the original authors, and also hoping to promote more meaningful and valuable collaborations.
If you are interested in sharing your latest work with more people through 【Deep Blue AI】, please click on the tweet below for details👇
【Deep Blue AI】Original content is crafted with care by the author team. We hope everyone adheres to the rules of originality and cherishes the authors’ hard work. For reprints, please privately message the backend for authorization, and be sure to indicate the source as 【Deep Blue AI】 when publishing, otherwise, legal action will be taken for infringement.
*Click to view, save, and recommend this article*