Comment by metalliqaz

given that LLMs are going to produce code that is essentially an average of the code it has been trained on, which is all human code of varying quality, I don't see how the current methods are going to actually produce better code than humans do when working with their own domain-specific knowledge.