Now here's an area where one of those newfangled large language models actually could do some good: create a spelling checker which doesn't gleefully replace somewhat misspelled or 'mis-swiped' words with correctly spelled but contextually clearly nonsensical ones.