To be honest, I have no personal experience with LLM (kind of boring, if you ask me). I know do have two collegues at work who tried them. One -who has very basic coding skills (dixit himself) - is very happy. The other -who has much more coding experience- says that his test show they are only good at very basic problems. Once things become more complex, they fail very quickly.
I just fear that, the result could be that -if LLMs can be used to provide same code of any project- open-source project will spend even less time writing documentation (“the boring work”)
As a sidenote. This reminds me of a discussion I haver every so often on “tools that make things to easy”.
There is something I call "the arduino effect:. People who write code for things, based on example-code they find left and right, and all kind of libraries they mix together. It all works … for as long as it works. The problem is what happens if things do not work.
I once helped out somebody who had an issue with a simple project: he: “I don’t understand it. I have this sensor, and this library… and it works. Then I have this 433 MHz radio-module with that library and that also works. But when I use them together. It doesn’t work”| me: what have you tried? he: well, looked at the libraries. They all are all. Reinstalled all the software. It’s that neither me: could it be that these two boards use the same hardware interrupt or the same timer he: the what ???
I see simular issues with other platforms. GNU Radio is a another nice example. People mix blocks without knowing what exactly they do.
As said, this is all very nice, as long as it works
I wonder if programming-code generated by LLMs will not result in the same kind of problems. people who do not have the background knowledge needed to troubleshoot issues once problems become more complex.
(Just a thought / question … not an assumpion)