I can only type fast with a lot of typos, typos so bad that the edit distance between what I want to say and what I write is too big for autocomplete.
On the other hand LLMs can clean up my jumbled typing if I paste it into them, say, after every paragraph. So what I would want is something like a local LLM fine-tuned for cleaning my typo-ridden paragraphs every time I hit Enter. Maybe it could also correct some grammatical mistakes like doubled words ("e.g. "the the"). The two places I feel I would need it the most are LLM chat interfaces (I know I could have a base prompt like: "always rewrite my input without typos before answering", but it costs tokens and time) and messaging (Teams, Slack, email)
I could not find anything like this while searching online, I found these but both are based on shortcut expansion:
https://news.ycombinator.com/item?id=37326870
https://news.ycombinator.com/item?id=24753493
What do you use for faster input?
I know of the options of learning to type faster and using voice input
I haven't seen anything like your LLM cleanup idea, but I like it. Why not implement it yourself? A MVP could take your clipboard, clean it up, then dump the cleaned result into your clipboard.
In the long term you would want to fine tune a small model, but in the short term just call openai/anthropic's smaller models with a terse prompt.
Also... wouldn't it be helpful to be able to type faster, regardless? It's a skill that more than pays for itself over a lifetime/career.