[WillAdams]

My spin on this is that if there isn't a large corpus of code addressing the problem, then the LLM will do poorly for want of training data.