Predictive texting/typing was basically the use for language models before LLMs. Not all language models are LLMs. And yes, correcting/predicting your typing has been the main feature for at least the last decade.