According to the scientists, there’s no universal method by which data can be deleted from a pretrained large language model.