Ask HN: Codegen Quality Is Interesting

I've been using LLMs to generate and edit code a lot lately, noticed a number of recurring issues (e.g. conditional errors causing incorrect behaviours, incorrect logical directions being generated, and in some instances code just not compiling).

Curious if anyone using open source LLMs to generate code has experience similar issues, and how they've worked around them without straight up fine-tuning their own models.