Given that chat gpt is trained on the internet. Places like Reddit. Think to how many times you have seen code written with the caveat 'this is a general gist.

Given that chat gpt is trained on the internet. Places like Reddit. Think to how many times you have seen code written with the caveat 'this is a general gist. I haven't tested this code' I've seen it many times. More than anything else really. So it should come as no surprise that models trained on that data produce similarly wonky code. This is then compounded by the fact that the user thinks of the model as they would a person. They receive a humanesqe response. So attribute the expected human backend. The understanding of the request and the feedback. All current "AI" LLMs are just Chinese rooms.