Not sure if you're violating rule 4, but no, assuming you're talking about current AI models. LLMs work in tokens, not binary.
i tried once ai with assembly to build an app from scratch; it wasn't time/token-efficient at all.. binary, it'd be worse i assume