Probably it is also depends on how much information about “various models trying to naively backstab their own creators” there are in the training dataset
Probably it is also depends on how much information about “various models trying to naively backstab their own creators” there are in the training dataset