To be strictly correct, it is always an approximation, and can easily diverge rapidly from the higher level. Literally the exact same situation as floating point numbers from double to single precision.