I believe the mistake to be yours. Algorithms is a largely theoretical field - when an algorithms researcher comes up with an algorithm with a lower asymptotic bound, that's all he or she is claiming - they're making no statements about its practical utility or the performance of an implementation. An implementation will almost certainly be provided, but that's besides the point.
An explanation of why it's useful in practice is again totally besides the point - and really would be taken to be rather insulting by many researchers - the motivations are often largely theoretical.
As a case in point, the fibonacci queue noted by OP was motivated by reducing the theoretical worst-case asymptotic complexity of Dijkstra's algorithm, and does that very - no benefit intended at all for normal programmers.
If people do not know what big-O means, that's their own problem. The algorithms people do their research, and their papers are published in the context of an informed reader - if someone does not know the definition of big-O and make mistakes based on it, the time wasted by them is probably their own fault.
In the case of this article, it seems as though the code originally ran in 200ms. This seems like a very acceptable amount - and one that would be improved more by using an algorithm like A* with decent heuristics than by simply trying to do the same work in less time.
An explanation of why it's useful in practice is again totally besides the point - and really would be taken to be rather insulting by many researchers - the motivations are often largely theoretical.
As a case in point, the fibonacci queue noted by OP was motivated by reducing the theoretical worst-case asymptotic complexity of Dijkstra's algorithm, and does that very - no benefit intended at all for normal programmers.
If people do not know what big-O means, that's their own problem. The algorithms people do their research, and their papers are published in the context of an informed reader - if someone does not know the definition of big-O and make mistakes based on it, the time wasted by them is probably their own fault.
In the case of this article, it seems as though the code originally ran in 200ms. This seems like a very acceptable amount - and one that would be improved more by using an algorithm like A* with decent heuristics than by simply trying to do the same work in less time.