The higher the base of a number system then the fewer digits it would take to write very large numbers right? So what if you counted numerals,lowercase and uppercase letters, and the emojis all as digits for a base300 system or whatever. Likely less than the full set of options to eliminate any visually ambiguous characters in the set. Then could you compress a file by writing it to this base and printing it to text file that you could then pass to an AI to visually identify each glyph and convert back to the binary string as it goes? Like reading an analog record?