I would love to have a unicode character for representing the start/end of a data block so that LLMs could at least send data meant to be uncorrupted down a different path at tokenization.