There is nothing about LLMs that make them bias towards "better" code. LLMs are every bit as good at making low effort reddit posts and writing essays for Harper's Magazine. In fact, there's a lot more shit reddit posts (and horrible student assignment github repos) than there are Harper's Magazine articles.

The only thing standing between your LLM and bad code is the quality of the prompt (including context and the hiddem OEM prompt).