Comments 3
Well, you are driving a conclusion from one evidence and just a bit of practice. In reality, there are many tasks it cannot handle, but many tasks it can. Also many ways to make it not work, and many ways to make work. No one tells it writes working code for you. It's just your slave author. What you do with the result is completely up to you. In my case – saving hours and money for the bulk code, especially in areas where I don't have deep expertise and had to pay other people previously to create such code. The code isn't for copy-paste for sure. But usually it's just a bit of correction and good to go, for an expert engineer it's a real time saver.
" No one tells it writes working code for you " - that is correct, but this is what many hope to get (and often do), me included :-)
I do not question LLM usefulness when used carefully, they are indeed can save alot of time.
I think it's quite far from writing code like this, and it is only good for short enough codes. It's a kind of a jump-start tool. It closes the gap between 0 and 1 in quite many areas. I find it very useful in automation like scripting and ops, where the complexity is not logical but rather technical. You don't know and don't want to know all the intricacies of bash and docker, but it does. Yet, in algorithms development... At current stage this would certainly not replace humans.
Doing 10 minute task in 2 hours using ChatGPT