I've been using gpt4 for some time now (I have the dev invite and subscribe to gptChat +). I am a softwr engineer with 20+ years of experience. Some things GPT outputs are really good and feel magic. However, anything slightly more advanced (as in coding) it is bad, like really really bad, not even junior level programmer bad, but much worse.

I highly recommend the paper that came out from MSFT last week "Sparks of intelligence of GPT4" (or something like that) It does a great analysis.

This is my approach on using GPT: If the task at hand is of low cognitive effort for me but tedious, I get GPT to do it. If the task would be hard for me to do (as in you need to take out a pen and paper and doodle stuff), I won't even dare to ask GPt to do it, it will be nothing by disappointment and more importantly, wasted time. So I'll do that one myself.