As long as LLMs are trained on code that has bugs, they'll suggest code that has bugs.