Hacker News Reader

Comment by jtietema | Hacker News Reader

Hasn't Firefox been doing this for a while already with their local translate feature?

https://www.mozilla.org/en-US/firefox/features/translate/

Maybe they mean LLM, instead of AI (not sure if Firefox translate models are LLM ir not)

Yes. I was responsible for kicking off that project. At Mozilla we'd been working with Europe on the Bergamot LLM translation tech and in 2022 we started work integrating it into Firefox. It began shipping last year. New language models are added regularly, all pivoting off English so French to German would actually be French -> English -> German and so you'd need two models for that. Each model was a few megabytes when I was starting on this but I've been away for a bit now so maybe they've improved on that but it did mean that we had to pull models down on demand because doubling the binary size for an initial Firefox download wasn't an option. (BTW, I was the guy that shipped Firefox v1 and my target for its download size back then was 3.5 MB or "about the size of an mp3". We made it down to about 4.5 MB which was close enough. It's over 10 times that size today.)

They use neural machine translation [0] models, they are artificial neural networks, but I don't think that you can really call them "AI".

https://en.wikipedia.org/wiki/Neural_machine_translation

See also:

https://browser.mt/

https://blog.mozilla.org/en/mozilla/local-translation-add-on...

Firefox's translation feature was built on Mozilla's Bergamot collab with Europe. It's an NMT as you note but it's also a set of language models trained on large language corpora using, among other techniques, the transformer architecture, so people calling it LLM translation don't bother me at all, and I'm the guy that brought it to Firefox (with my architect partner Andre Natal). The best article is probably https://hacks.mozilla.org/2022/06/training-efficient-neural-... though that was before we started integrating it into Firefox proper.

Thanks for the work you do, and your reply, very interesting!

Reply on news.ycombinator.com