@DaveMWilburn It's a beautiful job of "trainingwashing": because of the nondeterministic brute-force method of data compression, they can frame their "storing stolen data in searchable, highly compressed form" as if it wasn't "storing". But it clearly is, and done in a way that only people rich enough to somehow run a 900b parameter model as t heir daily driver can see that 100% of the data is there, clear as day.

Oh, and packaged with a nice language translation engine. A nice touch.