To me, this was a game changing update. The code that v4 was outputting wasn’t always perfect, but it was light years ahead of 3.5. As an experienced but not formally trained coder, I can usually identify issues or details that the AI missed. But overall, the code it generates is useful and easy to refine. ChatGPT v4 is even more helpful for the planning stages of a coding project. It’s also pretty good at debugging code.
One of the major limitations is simply how much information you can send or receive at one time. This makes it great for working on small projects and individual pieces of a bigger application. However, it’s not able to generate or debug a full-fledged application. There are simply too many characters for the AI to focus on at one time. Like us, but different, it has limited capacity for “short-term memory.”
I predict that this limitation will become less and less of a factor over time, and there are already many techniques for reducing its impact.