You can customize it, but I’m not sure what the default is for new users. It would be interesting to check.
As of this writing, there are 66 votes on the first question in the survey. Let’s do some analysis! Perhaps the simplest form of analysis is reading the results to “which policy would you prefer?” directly. In that case, “Policy 2 - Allow LLMs for translation” is winning, with 47% of people preferring it.
A different form of analysis we can do is ranked choice voting. For this, we take all of the poor people who voted for “Policy 1 - Ban LLMs for translation” and reassign their vote to what it seems their second most preferred option is.
9 people appear to like policy 3 second best, 1 person appears to like policy 2 second best, and 2 people weren’t able to be counted due to providing insufficient information or being equally in favor of each.
Details
ArDrakho - Didn’t specify which one they preferred least, but is not so strongly against policy 3 as policy 2
MistressRemilia, PulkoMandy, cafeina, naemuti, nipos, x86k, zeldekatze, _AP - Specified policy 2 as their least favorite and so policy 3 is likely their second favorite
Pap - Specified policy 3 as their least favorite and so policy 2 is likely their second favorite
us3r1d - Didn’t specify which one they preferred least and was equally against both options, so their vote will be simulated as split between the two, which means it cancels itself out, which means I will be ignoring it.
walaki - didn’t specify enough information to figure out their second-favorite option.
Adding all of the votes, it results in a tie with 32 people supporting allowing LLMs for translation and 32 people supporting LLMs for translation, but encouraging non-LLM methods. Ironically, the last time I performed this analysis was around 55 votes and it also yielded a tie back then, so they seem to have been really close for some time.
Now, is ranked choice voting the correct way to interpret these results? Who knows! I suspect there are many other good ways of analyzing the survey. But now you know that it’s a tie between policy 2 and policy 3 currently under simulated ranked choice voting 
Provided it stays with “encouraged”, I am prepared to change my vote from 2 to 3.
Well, I’m just one moderator out of a team of moderators and a relatively inexperienced one at that, so I can’t give any guarantees. But given opinions are split between 2 and 3, I think it’s quite unlikely that we go with “LLMs for translation are allowed but discouraged.” If the vote were split between 1 and 3 instead, then it would be much more likely that we would go with such a wording.
Making of that what you will
1 Like
How interestingly you turned everything upside down.
Главное не как проголосуют, главное как посчитают.
Это крылатое выражение, чаще всего звучащее как «Неважно, как проголосуют, а важно, как посчитают», обычно приписывают Иосифу Сталину в контексте выборов.Историки сходятся во мнении, что сам он эту фразу никогда не произносил. Впервые нечто подобное появилось в мемуарах бывшего советского дипломата Бориса Бажанова в 1928 году, а широкую известность среди русскоязычной аудитории цитата получила благодаря роману Анатолия Рыбакова «Дети Арбата».Смысл выражения лежит в плоскости политической сатиры и означает, что при недемократических процедурах исход голосования предопределен административным ресурсом тех, кто контролирует подсчет результатов.
2 Likes
Even if that may be outdated, as we can see on users map, Haiku has users literally all over the globe. I would be surprised if a translator was existing for each and every language. If a small country is a market to take or has a strategical position on the map perhaps but otherwise… So, unfortunately some may not have a choice and may have to pick the only translator existing. Others more unlucky may have to use a chain of two translators. Living in countries that are using languages widely spoken, we tend to forget about that but to consider when you’re voting.
1 Like
I treat Policy 2 and 3 as basically the same thing. Encouraging classical translations (while LLMs are still allowed) pretty much means most people will interpret it as LLMs are fully accepted, ignoring the “encouraging” part, which is just a suggestion.
The only way to slain the Lernaean Hydra is to cut all the heads and cauterize them. If you start making compromises just to get along, you are already sold to the devil.
For me at least, Policy 1 is the only option. Yes, some people will ignore that too, and just use LLMs anyway — complete with all the “hallucinations”, environment pollution, and zero learning in the process. But the same people will ignore any policy that don’t like anyway.
1 Like
I honestly found this survey to be difficult to answer as I don’t want to just ban something which enables people to communicate. However, I also don’t want AI boosters having a foot in the door by allowing LLM usage on the forum either. As has been pointed out, a lot of traditional translation services have turned over to using LLMs in the background aswell, which would make translating stuff without an LLM difficult. For self-hosted solutions (where it would then be possible to ensure that it won’t switch over to using LLMs in the background) I found libretranslate. As a personal experiment I’ve set up an instance on my tiny VPS under https://translate.zeldakatze.de (please note if you want to try it out that it prints out any request (including the input text) that’s being sent to it on the console, which would allow me to see it if I would look there). I make no gurantees that it will stay up for long (or for how long), as it is quite compute-intensive relative to my cheap VPS and it also has some other tasks which are quite essential to me. Also I set the character count to 1000 max to limit the resource usage, please don’t spam it.
Translating the above text (Till the point where it cuts off after 1000 chars) from english to german took around 20 seconds. I have no clue how long an LLM running on that VPS would take, but I doubt that 2 cores and 2 GB of RAM that are shared with other services would make a pleasant experience, if it ran at all.
2 Likes
LibreTranslate is really a nice tool,its results aren’t perfect but good enough to understand the meaning of the text,and it’s open-source.
I’ve known it for a few years already,but I somehow assumed that there would be much more public servers running it.
After searching a very long time with multiple methods,I found only one other publicly available server that works:
If there’s more interest for it,I guess I can also setup a LibreTranslate instance on one of my dedicated servers.
As the code can be also run locally,it would be nice to have native Haiku GUI application for LibreTranslate that doesn’t need an external server.
Well,maybe some day when I have time and finished all the other stuff… 
2 Likes
They already have more than a foot in he door as promoting vibe coded and AI assisted apps is allowed. I think eventually the forum will collapse under a flood of low quality apps.
It was nice chatting with all of you for a while!
4 Likes
Indeed, the category was more intended so people can continue talkiing about some stuff that now includes some generated code. Not so people can send endless changelogs of vibe coded apps with 0 engagement.
2 Likes
What’s the problem? Just ignore the post under that category
4 Likes
I do not think that it will happen, Haiku is not so popular that everyone will rush making applications for it using AI agents. No need to solve problems that not occurred yet and may never occur.
7 Likes
Allow translations, just make text reviews for avoid mistakes. Let someone who understands the language handle it. What are you guys waiting for? If not, just leave it for review if you want to avoid mistakes. Or just leave it as is if there’s no native speaker available. At worst, it’s no big deal if an incorrect translation slips through. Later, if any errors pop up, they’ll let you know. This isn’t code generation, which, if left unchecked, can seriously mess up the codebase, and that’s usually a pain to clean up. It’s actually much simpler than you think.