When attempted to translate supposedly from "Belarussian" into "Russian" language, AI translation model intentionally adds comma even though there wasn't one in the original message.
As we all know from morality tales one comma may turn the sentence's meaning up a notch
In my opinion, would be a good idea to prevent this sorts of things from happening
thinking of adding more precise system prompt for models?