Danielle ForéUghhhhh someone submitted an app to appcenter that is not only obviously vibe coded because Claud is on the repo but also because there’s a lot of really big mistakes.
Honestly I don’t like this at all. If you want to learn to write apps please ask for mentorship and join our community there’s lot of people willing to help. But I can’t help but feel like this is a waste of my time because you’re not learning anything this way. It feels gross
I also feel like we need something in appstream so that people in the App Store can know an app was made with an LLM and avoid it
GeopJr@danirabbit was also thinking that but couldn't decide between suggesting a freeform "ai disclosure" or an OARS-like approach so we can list 'code', 'translations', 'branding'...
Victoria🏳️⚧️@GeopJr @danirabbit for translations, would using tools like google translate also fell under AI usage?
@tragivictoria @danirabbit for me it depends. The harmful part of AI translations is developers straight up translating their strings into 100 languages they have 0 knowledge on and calling it a day. In that case marking it as such would help set some expectations that the translation quality will be bad (and should avoid installing it on your non-English speaking relative's computers).
Now if translators use AI themselves to speed it up it's 🤷 for me. In fact many translation platforms do that automatically as suggestions.
Now if translators use AI themselves to speed it up it's 🤷 for me. In fact many translation platforms do that automatically as suggestions.
In fact many translation platforms do that automatically as suggestions.
Can I infodump how Pontoon handles it?
@tragivictoria sure! (let's untag Danielle)
@GeopJr so Pontoon does indeed shows AI suggestions to translators, and by clicking on them they get copied to text area for translator to edit or click straight "submit". Its done using Google Cloud API and they cannot be disabled. However, in the hamburger menu there is an option to make the AI suggested text more informal/formal, and that is done using OpenAI API and Mozilla is paying for its usage.
There are also pre-translations, which are done using custom language models trained on language team's approved strings (i.e. strings that were pushed to the main), however it's opt-in for locale teams, and my team wasnt using them.
Personally, i only ended up using AI suggestions either same way as I would Google translate (i.e. when im stuck), or as a way to save up on typing (i have weak finger joints). I never used the "make more informal/formal" option.