Pretty much the same, when I try to get a LLM output correct code targeting a certain libary, but in its training data are various conflicting versions of the libary and the result is a incompatible mix composed of code for different versions thrown together.