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?