Hacker News new | past | comments | ask | show | jobs | submit login




I brought up translation as a risk with a friend. If you pay someone for a translation these days, there is a chance they will just feed it to some AI to cut costs. You'll have no way to validate yourself if you don't speak the language.


Do you usually just pick a person at random when hiring or do you spend some effort looking into their qualifications and references?

How's your hypothetical any different now than it would've been in the past 15 or so years of Google Translate's existence?


Google translate is nowhere near as good as GPT4 at translation. Especially when given additional context and style instructions.


Sure, but people get burned despite attempting to be careful all the time.

Translation software predates Google translate, and I'm not claiming something has suddenly changed. It's slowly gotten better, and I assume the temptation to only pretend to have human translation will keep growing with it. The safer a scam seems, the more people will try it.


The shape of managing the work approaches the work in terms of fractal complexity


You do, actually: feed it to GPT-4 several times in different sessions. When it hallucinates, translations come out obviously different. When it actually knows what it's talking about, they'll match except for minor things like word order and synonyms.


Just a chance? I routinely translate hundred pages of pdfs to greek, in 3 minutes. The translation is far from perfect depending the text and it still needs a human in the loop for corrections, but i couldn't imagine translating a 300 pages pdf to greek by hand.

There is also the translaxy bot on poe.com which i use to translate english or modern greek to ancient greek. Out of this world good translation.

I mean, are humans still employed to translate text? Like an employee doing that job, and only that?


Hundreds of millions are spent on translators every year. It's a major expense in the EU budget, for example. A lot of people are going to jail for fraud if people aren't actually doing the work.


Oh, didn't know about that! Learning something new everyday i guess. Automatic translation works very well for technical documents, but it doesn't work that well for novels. So i thought, most of the translation jobs would be gone already. I think, given a little bit of time, a handful of years, translation will be automated 95% or more, across the board, for every kind of document.


I know people in the translation business. Automatic translation is out of the question in a lot of businesses and areas. One category would be safety-critical businesses where companies can get into a world of legal hurt if their translations aren't exact (medical, legal, defense, etc.) Saving a few bucks on automatic translations loses its appeal if you really need the text to be correct. The translators will also be liable for wrong translations, also a very important factor for professional clients.

Another example would be highly specialised, industry-specific texts with a lot of jargon, for which professional translation offices get you translators who aren't just fluent in the languages, but also knowledgeable about the area the text is about.


Have you seen any good translation tool from video to text yet? I'm trying to find something for Estonian to English and having little luck


Do you mean images in the video like how the Google Translate app can do with the camera, or do you mean the audio within the video?


The audio within the video


Unfortunately, none that I'm aware of. For whatever reason, I find that speech to text is never as good as the accuracy scores claimed by those making the models.


Whisper does audio to text with translation, so you'd have to extract the audio channel first.


This is actually a really good parallel.

Understanding the output of an LLM is similar to the output of a translater.

If the recepient doesn't/can't understand it, all bets are off.

Say you don't understand python but have an LLM write some for you, but you have no way of knowing what it's doing.

What if you have a malicious LLM hosted somewhere and it writes malware insatead of what you asked for.

If you don't understand the output you end up with, you run it and it pwns your network.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: