[brabel]

LLMs work better as translators than any non-AI translators though. Because they are able to translate not just words, but also capture the context of what's being said. If you translate a common phrase like "home, sweet home" to another language, it may or may not make any sense if you translate it word-by-word, like traditional translators would normally do... but LLMs know "what you mean" and will use the equivalent saying in the target language, even if that use entirely different words.

[cobbzilla]

I dunno? I think modern translators get idioms nowadays don’t they? If not, they should.

how hard is it to recognize common idioms and at least state the literal meaning followed by the semantic meaning? there are at most what, a few thousand per language?

[ozim]

I think someone who has low level of English will benefit more from trying to write on his own.

Unless they don’t care about learning English which shouldn’t be frowned upon.

[bmacho]

Yes, but also no. The properties of a style lie in how it is perceived, and LLM output style stinks as hell right now.

Google or Bing translate might not use the exact same words and phrases that LLMs use every single time, so you are better off using those

[watwut]

Human translators did not translated word for word. That part is simply untrue.

And LLM does not know context, it makes mistakes a lot more in it. But, it is much cheaper.

[Xfx7028]

I think he meant non-human translators, like Google translate etc. Those translations were indeed not making any sense sometimes. Although I have heard that they improved Google translate in the recent years.