Why Artificial Intelligence is a Better Programming Teacher than Humans

"Copilot greatly speeds up my coding speed. It's hard to imagine going back to the era of manual coding," wrote Andrej Karpathy. While there are many programming tools to choose from, GitHub Copilot has become the bible for programmers. This tool has been used by over 1.5 million developers of all kinds, and some believe it makes programmers "lazy and incompetent." Unlike Copilot, which allows users to make requests and have others complete them, Stack Overflow requires understanding of basic principles and finding the correct methods through trial and error. Meanwhile, Code Whisperer recently released a customized update that further simplifies this process by generating specific code based on their private libraries, allowing users to simply verify and check the generated code. These tools can be seen as advancing to improve work efficiency, similar to the transition from paper maps to smartphone maps. "Computer Education in the Age of Artificial Intelligence" In a study conducted at the University of San Diego, professors from nine countries were interviewed to understand how they are adjusting their courses to accommodate the increasing use of AI coding assistance tools, such as ChatGPT and GitHub Copilot, by students. In the short term, they are particularly concerned about cheating. The authors of the paper, Sam Lau and Philip Guo, wrote, "Professors are concerned that excessive reliance on AI tools may hinder students' comprehensive understanding of programming fundamentals, similar to using a calculator without understanding decimal points in math class." The heavy reliance on these tools has led users to flock to social media to complain when they were unable to work when ChatGPT was offline last month. Initially, there were concerns that Copilot would take over the work of programmers, but it was quickly realized that it is simply a tool that can automatically correct and improve the efficiency of developers' work. Nevertheless, there is shocking evidence that AI tools are better at helping students learn programming better and faster than teachers. A study conducted at the University of Toronto examined students learning programming. They found that AI programming assistants like OpenAI Codex can help novice programmers write code faster and better, while reducing frustration, without compromising their performance in manually modifying code without AI assistance. David J Malan's popular CS50 course announced in June that they are integrating AI grading and teaching coding together. Not only Harvard University, but other Ivy League schools are also adopting AI to improve student performance while making it easier for teachers. Meanwhile, Stanford University has developed an AI teaching tool that provides feedback on homework for students. Assistant Professor Chris Piech said, "One thing we can't really do is expand the feedback loop. We can expand instructions and content, but we can't really expand the feedback loop." "Fortune and Misfortune Coexist?" Like any technology, there are still some issues with AI coding assistant tools. Some are concerned that they may create and reinforce winner-takes-all situations. Since few companies have the data required to build such tools, creating competitors to Copilot will be challenging. Recently, Zoho announced the development of a similar "programmer productivity" platform for code generation. GitHub Copilot has an accuracy rate of about 26%, which is very poor if users don't have a good understanding of the basic concepts. Although ChatGPT has higher accuracy, it may contain subtle errors that beginners may miss. These tools are useful for learning and experimentation, as one Reddit user said, "I spent 30 minutes just asking it 'What does this do?' 'Why did you do that?' 'Why not do this?'... It's like an experienced programmer explaining everything." Another benefit is that using AI tools makes it easier to learn new programming languages after becoming familiar with one.