Comment by Markoff

Some of my clients use the LLM to do QA on my translations, dunno about English, but for smaller languages (think less than 10M speakers) LLM still ain't that good, at least 5-10% LLM suggestions contain hallucinated made-up words which don't even exist.

I can see LLM usable for long paragraphs of text, legal documents, maybe even books if you don't care about artistic quality, but good luck translating software/UI with short strings with hardly any description/screenshots even human has trouble to figure out let alone dumb intelligence.

I already told one of my clients who work for one of the major Chinese companies I can't guarantee quality of translation if they don't provide me variables description and all screenshots are in Chinese, so basically you are blind whether you are translating button, toast with instruction for user or page title, but seems client doesn't care...

Nowadays I don't even bother explaining why are AI reports wrong, it would be huge waste of my time, if 99% are FP.