The typos are from the OCR. It can't read OP's shitty font on top of the images.

(Still theoretically possible for language models to make "typos" because they exist in the training data, but very unlikely )