Make ChatGPT Your Intern, Not Your Boss
How I used ChatGPT to solve a real-world coding problem
If you’re a software engineer trying to use ChatGPT to code, you’re probably doing it wrong.
You’ve heard about the incredible applications of ChatGPT in education, law and healthcare, to name a few, and as a result you might expect it to write amazing code. And to be fair, ChatGPT does often solve simple coding challenges well. But as soon as you try to solve a real-world problem, it breaks down — ChatGPT either doesn’t understand the problem correctly, or worse, produces code with subtle bugs that can be very difficult to detect and fix.
The problem is that ChatGPT isn’t really understanding your question and thinking about it in the way an engineer would. Instead, it’s pattern matching your prompt against a huge corpus of training data. ChatGPT behaves more like an intern who needs to be coached to a solution than a senior engineer who can solve problems independently.
We have to adapt our approach to make use of this technology. Instead of treating ChatGPT like an oracle that answers hard questions, treat it like an intern that can tackle small, well-specified tasks. And expect to have to go back-and-forth with it before getting to the right solution.
