Comment by dandersch

> Small quantities of poisoned training data can significantly damage a language model.

Is this still accurate?