I have noticed LLMs tend to generate very verbose code. What an average human might do in 10 LoC, LLMs will stretch that to 50-60 lines. Sometimes with comments on every line. That can make it hard to see those bugs.