All right, we’ve discussed (and will probably continue to discuss) philosophy, hallucinations, quality, personal freedom, and more. I want your thoughts on the various policies we might adopt.
Policy 1 - Against LLMs for translation
The rule would state to not use LLMs for translation. As justification, it might point out the potential for hallucination, environmental impacts, and lack of transparency into the inner workings. If somebody uses LLMs without telling us, we would have a hard time knowing. If, however, the LLM hallucinates, they would either have to pretend that what the LLM said is what they meant or admit they were using an LLM.
Policy 2 - For LLMs for translation
The rule against using LLMs for forum posts would gain an exception: you may only use LLMs to translate. It may also advise you to check the translation for hallucinations by either reviewing the translation yourself or using a classical translator to translate the message back into your language to check for accuracy. Alternatively, it might require putting the original untranslated message with the translated message so that in the event of misunderstanding we know what you meant to say.
Policy 3 - Compromise
It would still allow LLMs for translation like policy 2, but encourage you to consider using a classical translator instead because of the directness of the translation, environmental cost, and similar.
If you have any other compromise solutions you can think of, I’m interested. This is the only one I could think of.
Yeah, if your reply only rehashes what’s been said before, I would ask you to consider whether that would actually add anything to the conversation or just make the conversation go in circles.
Yes, I’m thinking of doing some polls later, maybe after a day or two of discussion, to judge what people think of the various policies
OK, so how about we agree that “generated” in the phrase “generated by large language models” in the policies does not refer to LLM-backed translations services and then add a section like:
English is the most commonly used language here, but not everyone can read or write in English. Use the “International” category to post in languages other than English.
If you use an automated translation system to participate, you can add your original text in a collapsed box to allow readers of that language to see your original intent. This isn’t required, but moderation decisions can only be based on what is actually posted so we encourage you to do this.
This isn’t ideal from everyone’s perspective, but I think there’s very widespread support for not officially caring what backend people’s translators use and I personally want to reinforce the policy that “the translator added that” is not an excuse.
I don’t like this part, it sounds like “you better do this, or you might get banned", and goes against the presumption of innocense, in the first place I would assume users want to be nice and acomodating, and not willfully violate rules and make sure there is an “auditability”. If there is a problem we can just talk it out, and in general wo only ban after repeated infractions anyway..
I was trying to get an explicit statement into it that “the translator did it” would not be accepted as an excuse for a policy violation, which I think is an important side-effect of the poster handling the translation.
There are some unwarranted assumptions here: that the user actually knows what back-end their favourite translation software uses, and that “classical” translation sites will still be around in the future. Neither of those is a given.
We have seen some people agreeing to paste their original text, and others strongly disagreeing. So I don’t see how this can be more than a strong recommendation
rther than a requirement.
^^
[Compulsory spelling error to prove human origin]
I don’t think that would be reasonable in any case, if the user was able to perfectly interpret the translator they would not need a translator. We can of course tell someone who repeatedly created confusion with this that this is not acceptable, butt hat would be something afterwards
Well, not necessarily: I can interpret english, also when sentences are complex, but I’m not able to write those complex sentences myself without a translator. Moreover, using a translator lifts me from the frustration of not being able to fully express my thoughs correctly
All right, it seems discussion about LLMs for translation has mostly died down. Let’s see how the community at large feels about each of the proposed policies.
First, which policy would you prefer?
Policy 1 - Ban LLMs for translation
Policy 2 - Allow LLMs for translation
Policy 3 - Allow LLMs for translation, but encourage non-LLM methods.
0voters
Second, which policy do you like the least?
Policy 1 - Ban LLMs for translation
Policy 2 - Allow LLMs for translation
Policy 3 - Allow LLMs for translation, but encourage non-LLM methods.
0voters
Now, how much do you like/dislike policy 1, which bans LLMs for translation?
Strongly against
Against
Weakly against
Neutral
Weakly for
For
Strongly for
0voters
How much do you like/dislike policy 2, which allows LLMs for translation?
Strongly against
Against
Weakly against
Neutral
Weakly for
For
Strongly for
0voters
How much do you like/dislike policy 3, which allows LLMs for translation, but encourages non-LLM methods?
