Anyone tried talking to GPT in a Slavic language? My experience is that it in general it can talk in Slovak, but sometimes it uses words that seem to be from other Slavic languages. I think, either it depends on how much input it had from each language and there are relatively few Slovak texts online compared to other languages, or the Slavic languages are just too similar to each other (some words are the same in multiple languages) that GPT has a problem remembering the exact boundary between them. Does anyone know more about this?
I get especially silly results when I ask (in Slovak) “Could you please write me a few Slovak proverbs?” In GPT-3.5, only one out of ten examples is correct. (I suspect that some of the “proverbs” are mis-translations from other languages, and some are pure hallucinations.)
Anyone tried talking to GPT in a Slavic language? My experience is that it in general it can talk in Slovak, but sometimes it uses words that seem to be from other Slavic languages. I think, either it depends on how much input it had from each language and there are relatively few Slovak texts online compared to other languages, or the Slavic languages are just too similar to each other (some words are the same in multiple languages) that GPT has a problem remembering the exact boundary between them. Does anyone know more about this?
I get especially silly results when I ask (in Slovak) “Could you please write me a few Slovak proverbs?” In GPT-3.5, only one out of ten examples is correct. (I suspect that some of the “proverbs” are mis-translations from other languages, and some are pure hallucinations.)