[avian]

> ML revolutionized translation a long time ago and the demand and pay for translators went down.

ML translation also caused a noticeable drop in the average quality of translated text in my experience. Companies now ship machine-translated manuals for minor languages that are often little more than gibberish. Human translators weren't perfect - for example, often you could see that the translator didn't know the terminology of the field. But at least the rest of the text was usually intelligible.

If this is where we're heading to with visual art as well I'm not looking forward to the future. Imagine an instruction manual where illustrations are machine-created. Everything looks kind of weird and inconsistent. If you quickly flip through the book it looks like all the important points are shown on the pictures, but looking closely shapes don't match with reality and the details are all wrong. The number of bolts changes from picture to picture. It's all there just to check a mark on someone's list, but doesn't really help you in servicing the thing.

[martopix]

I disagree with the first part, I remember home appliances made in Asia in the 90s that came with completely ridiculous manuals, possibly translated with a dictionary and little else. Machine translation can really be good if used properly. Of course I wouldn't use it to translate literature.

[bombcar]

It was better than those, but many "professional" companies that would in the past hire out a group of translators to translate their documentation into various languages now just use machine translators with a cursory pass.

If you read mainly English, you won't notice it, but you'll see it if you speak a language that does have translations, but aren't primary.

[vintermann]

That sort of translation always had a lemon problem. How do you know if the translation to Portuguese is any good, if you don't speak Portuguese and don't know anyone who does? You can pay the expensive translator company, or the dodgy cheaper translator company that claims to be equally good.

Did you choose the responsible expensive option? Sucks, because they have been competing with the bad/automated translators so long that they squeeze their human translators so hard, they produce bad translations anyway.

Also, it didn't help that you didn't give them any context, but just sent them one sentence at a time, extracted from the strings in your program. Because we've all done that, haven't we?

[MonkeyMalarky]

A DALL-E generated IKEA style manual for a nonexistent piece of furniture would be absolutely hilarious.

[merek]

Introducing the Fiibean wepa Taki: https://labs.openai.com/s/95bli4C4CKZ73DMigFNdDKQ4

[MonkeyMalarky]

Absolutely fantastic! You ask for four legs and receive five. I wonder what captain Picard would have to say about that.

[A4ET8a8uTh0]

Hah. Yes. And as an amusing result, emails and chats are filled with errors resulting from automated translation. The service was supposed to be a crutch, but it became widely adopted and lowered the standards everywhere.

Anecdotally, I now check my email, because Outlook likes to change what I wrote.

[shadowgovt]

It's honestly pretty fascinating to watch perfect be the enemy of the good in this space. As a result of the birth of cheap-to-free machine translation of human language, the space of things available on the global market has become incredibly vast, because the good-enough stab at translation is often good enough for a dedicated hacker to make the thing work anyway.

[ghaff]

It applies to even English language transcriptions as well. Do I pay pennies a minute or a $1+ a minute? It depends on the use case and my budget. If my time is expensive and I'm publishing externally (e.g. for a professional podcast or an interview for a client), I'll probably go with a human. If I just want to refer to some notes to pull out a quote or go back to check something? An ML transcription with timestamps is probably fine. Ditto if the alternative to ML transcription is nothing.