Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

How do the models do if you ask them to translate to X language and adapt the text to suit cultural norms and idioms?


At least in my experience poorly. As a reference I speak one of the smaller European languages and what is interesting that newer models tend to do worse, but what all of them struggle with and is very hard to control using prompting is tone.

Formal language in American English is very friendly, very sweet, to an extent that my language is not. Any piece of text therefore sounds like a psychopath is trying to manipulate you.




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

Search: