[imiric]

... That you're aware of.

The more code you ask it to generate, the higher the chances that it will introduce an issue. Even if the code compiles, subtle bugs can easily creep in. Not unlike a human programmer, you might say, but a human programmer wouldn't hallucinate APIs. LLMs make entirely unique types of errors, and do so confidently.

You really need to carefully review every single line, which sometimes takes more effort than just writing it yourself. I would be particularly wary of generated code for a dynamic language like Python.