This project does not allow contributions generated by large languages models (LLMs). We are taking these steps due to:
- The potential negative influence of AI generated content on quality, developer alienation from their code and skill rot.
- Legal complications such as the inability to claim copyright and ensure others’ licensing and copyright have not been violated.
- Ethical concerns including but not limited to those regarding intellectual property theft, environmental impact, the effect on already-marginalized groups, devaluing labor for the purpose of concentrating power among the billionaire class, etc.
This ban of AI generated content applies to all parts of the project, including, but not limited to, code, documentation, issues, and artworks. An exception applies for translating texts for issues, discussions, and their comments to English.
Previously, when AI-related discussions came up, I was using AI to translate my posts into English. I was concerned about becoming a target because of that, so I ended up closing all my issues and stopped participating in the chat.
It helps a lot that translation into English has now been explicitly allowed as an exception.
Out of curiosity, are non-AI tools like translate.google.com insufficient for your needs?
Edit: never mind. Apparently, Google translate uses Google‘s internally developed LLMs since 2020. :/
https://en.wikipedia.org/wiki/Google_Translate#Neural_machine_translation
Yeah, that’s definitely part of their internally developed LLM. I remember reading about Google developing this when I was in design school. I just didn’t know that Google translate was using it, too, but I suppose I should’ve known.
Thanks for posting this.
I use Google Translate when reading, but I still feel that Japanese translations are not always perfect.
For writing, I usually translate with ChatGPT, then check it again by retranslating into Japanese with DeepL, and pick the version that feels the most natural.
When I am translating from Japanese, I find that DeepL is the most consistently understandable and accurate. I have copy/pasted many an article from Comic Natalie into translators to compare with what Anime News Network writes, and that is my experience.
As someone that has English as second language, but Japanese not as my third, I like how grok AI translation works. Although confirming it with English speaking Japanese acquaintance they say while it is generally slightly better than google translate or deepL it is still wrong in a few ways often enough.
I say grok because there’s still a lot of Japanese artists (singing and drawing) and vtubers that still use twitter. It annoys me greatly but I understand why they still stay on Twitter for the audience, but especially for the drawing artists I don’t fully understand (or refuse to accept), twitter has repeatedly say they WILL steal their artwork to train grok.
If they jumped ship to misskey or something with a big Japanese audience on fediverse, I’d make it my mission to understand the micro blogging section of the fediverse lol
I’m more of a topic follower rather than people follower on fediverse so Lemmy, piefed, mbin is still my preferred home
I don’t particularly like ChatGPT. I tend to favor Claude just because it’s far less obsequious and tends to give a bit more pushback and is more critical.
I also like the structure of it answers a bit more. Maybe it’s just a matter of taste, but I really don’t trust an LLM model that keeps kissing my ass so much.