I’m reading that description and I don’t understand how this is different from literally every autocomplete/spellcheck function that learns to predict based on your past typing habits.
I thought the definition of LLM and why it’s different from other predictive text functions that have been on all sorts of devices for decades, is that it’s large and massively multi-victim and therefore not possible to localize to your device.
(I mean I do turn this shit off whenever I can, and I have for years, because it’s annoying. But I’m morbidly, dreadingly curious to know what’s new about this version of it)