PROBLEM
Nerwer AI models tend to answer meta-comments like "the translation is ambigous because this and that" or "The segment "Sth." could not be translated becaus ebla bla" instead of actully translating a segment.

SOLUTION
Amend our default promps by a instructon like

"Please do never answer with comments about the translation, always just return a translation or in the rare cases where no translation is possible just return the original segment"

Such comments can be challenged e.g. by very short segments that just represents a common abbrevation like "AI" or a product name lke "BMW"