Using a bad translation/transcript as base for professional translator is still better than nothing. Like I said, translators are still going to be needed, but lots of the heavy manual work can be now automated.

Also often when very domain specific language is used, the translation made by human can be bad, because they don’t know the proper terms. Of course good professional translators will ask these. It is also something that must be done with these dummy LLM models, you cannot just throw text into it and expect good results.