Bloom filters scale logarithmically with false positive rate and linearly with the number of items stored. The article doesn't mention the false negative rate.

Bloom filters scale logarithmically with false positive rate and linearly with the number of items stored. The article doesn't mention the false negative rate. Unlike a Bloom filter, it'll depend on order when the input includes repeated elements. But in general, required memory will increase quadratically in the number of items stored at a constant false negative rate (because of the "birthday paradox"). So it isn't the opposite of a Bloom filter. But what is?