current behemoth machines have errors too, often the best way to get a really polished sequence is to sequence on machines which have different type of errors, and interlace the data appropriately.