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.
Simon
OpenPGP_signature.asc
Description: OpenPGP digital signature
