On Sun, Jul 26, 2026 at 05:48:18AM +0900, Simon Richter wrote: > Hi, > > On 7/25/26 15:47, Matthias Urlichs wrote: > > > > I want to add, though, that I'm not sure how I feel about preventing > > > LLM usage for translation. > > > Seconded. This puts the onus of translating the content on the receiver, > > which (a) will result in far more LLM use (and thus waste of energy and > > all that) than if the sender did it, (b) marginalize the sender: not > > every intended recipient will bother. > > The problem with doing an LLM translation on the sender side is that if > there is information loss, it happens on the sender side, and cannot be > reversed by the recipient. > > Because of the way LLM based translation works, such information loss will > most likely take the form of "all the words you would expect are there, but > the meaning is different", which is difficult for a non-native speaker to > notice, so they would be sending a message with a different meaning than > they intended. > > I work in a bilingual company, and we have moved away from sender-provided > translations specifically for that reason -- if I do not understand a > sentence, I will usually try translation software first, but verify the > grammar against the original to find out if the translation is correct, or > the opposite of what was meant. > > The latter happens about twice a week, and while it is 面倒くさい to do all > of this translation work, it saves a lot of time compared to working on > incorrect specifications.
You already speak the language of the email at an intermediate (?) level. Most people in Debian might not be able to tell whether an email is written in Japanese or Korean, and would have to trust the translation blindly. People on a mailing list receiving different translations of varying quality and sometimes opposite meanings of the same email in a discussion - that would be worse than everyone receiving the same bad translation where someone might notice when something sounds mistranslated. > Simon cu Adrian
