Any topic with little coverage in the training data. LLMs will keep circling around the small bits in the training data, unable synthesize new connections. Thi

Any topic with little coverage in the training data. LLMs will keep circling around the small bits in the training data, unable synthesize new connections. This is very obvious when trying to use LLMs to modify scripts in vendor-specific languages that have not been widely documented and don't have many examples available. A seasoned programmer will easily recognize common patterns like if-else blocks and loops, but LLMs will get stuck and output gibberish.