Honestly, I'm both worried and excited. The excited part of me sees GPT-4 as a similar tool to Copilot where it can write concrete, well-defined blocks of code really easily, and that's actually great. Having it generate the CSS for a gradient background or build a simple modal saves me time and makes me as a developer more effective.
Today, I don't see it as something that can go build an entire real-world application with no direction. Even in this example, most blocks of code that it produced had at least one bug. And some of those bugs it just straight up couldn't fix.
The worried part of me assumes that it'll only get better with time. I'd guess that the most effective use of GPT-4 is going to be alongside a developer instead of a replacement for one, but who knows what future versions will bring.