And with machine code, you got to keep track of what's in the stack, CPU registers, ... to make a sense of what the code and the next branch command does. It's completely unalike processing human language. LLMs aren't really set up to do it.