They also point out that HLL doesn't have false positives but it does have incorrect counts (I don't really see how that's different tbh), and later say it has the same "absolute correctness" as an association table...