My guess is that it's roughly analogous to the difference between Wolfram and Google.

One is actually computing on the data to combine it in novel ways.

The other is the one I use.

(Snarky, but as much as I try, the only thing I've found WA useful is for COLA comps)