Text that, in the examples used to train the neural net, has next token targets that represent answers where the unprivileged text didn’t outsmart the privilege

Text that, in the examples used to train the neural net, has next token targets that represent answers where the unprivileged text didn’t outsmart the privileged text. But given that must over or underfit there is no guarantee that it will do perfectly well on test data at honouring this.