That's not necessarily the case in general. If you had an algorithm which, given a billion dollars' worth of specialized lattice reduction ASICs, could break 2

That's not necessarily the case in general. If you had an algorithm which, given a billion dollars' worth of specialized lattice reduction ASICs, could break 2048-bit RSA in a few months' time, then for all practical purposes this would mean that 2048-bit RSA was broken -- there are plenty of state actors who would drop that kind of money in a heartbeat -- but that doesn't mean that some guy who came up with it would be able to do a demo on his laptop. (Not saying that the above has anything to do with this paper in particular.)