Note, I don't have python3.4 handy, but on python2.7 this isn't always the case (and I'd expect that to be true of python3.4 too).
I can even change the order of the elements based on how I put them into the set:
>>> print({1, 9})
set([9, 1])
>>> print({9, 1})
set([1, 9])
>>> set([9, 1])
set([9, 1])
>>> set([1, 9])
set([1, 9])
The order is determined by the hash of the element and by when it was inserted (in the case of hash collisions). In CPython, integers hash to themselves and a dict/set has 8 slots free to start with. Since there are 8 spots available, we can hash numbers 0 -> 7 (inclusive) without a hash collision. However, if we try to hash 8 and 0 (or 9 and 1) in the same set, we'll get a collision. if 9 is already in the set and then we try to put 1 in, python looks and says "Oh snap, that slot's taken -- Now I need to put it in the next most favorable slot". The details of collision resolution are beyond what I've looked into, so I can't offer insight into what slot that is ...
Note if we had more than ~5 elements in the set, then it would be resized (IIRC, to 16, then 32, then 64, ...) which changes which elements would collide (naturally).
