I'm not surprised you weren't able to reproduce the mis-translations. Here on chesspub I only quoted some relevant snippets, but for the translation I was doing chunks of about 4000 characters at a time (Google has a limit of 5000). The extra context means Google "knows" the subject is chess and thus knows the most popular terms to substitute.

There is another hazard, which is that Google "knows" me and "knows" you, so we can get different results based solely on that, regardless of the inputs we provide. In a similar way it's harder than ever to get wide ranging query results from a browser search. The search engine keeps filtering the results for what it "knows" the user already wants. This was happening even before AI entered the picture. I often used startpage.com to compensate for google search personalization. Now I wish there were similar for other search engines.

Nothing wrong with ChessBase including defective technology in their product, as long as it's optional for the user. It would be unacceptable in a business context (although not all purchasers of business software would agree!), but for consumer software a great model is to get something/anything working quickly, then improve it over time.

Well, maybe we'll all finally able to beat Stockfish on level 10! (when A.I. assisted)