Comment by bjourne

The argument is that a modern LLM cares little for the MLE. Hence, statistically speaking, they are not predictors.