It means to change representation to less bits per number floating point, lower resolution numbers