“As with all language models, it is hard to predict in advance how language model will respond to particular prompts and offensive content may occur without warning. Depending on your usecase language model may produce socially unacceptable text: profanity, lewd, and otherwise abrasive language.”