Add to that the fact that matching parentheses is hard for humans to reliably do correctly—especially when typing code into a textbox on Stack Overflow—and now you have a model whose architecture struggles with knowing when to stop in highly repetitive sequences and that was trained on data that is disproportionately likely to have incorrect syntax for the language you're using.
Other languages don't have those features, and I've never found Copilot to give me a syntax error in, say, JavaScript.
That said, I also don't know any professional developer who trusts LLMs to code. Everyone I know who uses them uses them as a tool and takes responsibility for cleaning up their output.