A fascinating quirk of Large Language Models (LLMs) is the ability to accurately interpret input text that has been scrambled. This phenomenon has been documented and quantified in this paper arxiv.org.
To try this out for yourself, here is a quick tool to scramble the letters within a word.