Comment by usgroup

I think you’re wrong about the ML precision. You need highly precise for most recursive machine learning tasks because you’re compounding errors otherwise.

Typically you can’t even use floating point representation: not accurate enough.