Not really a suprise. I also noticed that AI results got worse lately. Like 1 year ago I was kind of getting decent results when asking to generate something but lately it’s pretty bad. And I don’t want to craft the right prompt and fiddle around for hours - then I can write the code also by myself.
I think the first AIs were trained on human-only generated code and the quality was quite ok. Now AI has already produced a significant quantity of less-than-ok quality code that humans spread across all the different code hosting sites.
That said quantity of AI-generated code "polluted" the training sets of all future AIs, because AI auto-feeding itself cannot give good results. I don't see how the quality can improve.
