"LLMs just predict the next token" Well what else could they do?! I don't think they'd impress anyone predicting the previous token