I’ve been using DeepL for a few months and love the translation quality but it’s language support is more restrictive than others.
I built something for myself today that uses DeepL when it’s best and GCP/AWS otherwise. I also made some efficiency improvements to use less input characters when possible.
Would anyone be interested in using this? I could probably offer it at cost but it’s not cheap.
I use the free version of Deepl practically every day, both for work (we have French customers) and for leisure (I use Nostr to write to you in English).
I think it is currently the top, but I don't think I will ever use its subscriptions.
I think so too. Mostly only companies will want/need to pay for access.
I seem to get better translation results from GPT models
I think it depends a lot on what you’re translating but I’ll do more experimenting!
i get random texts in chinese that i use deepl to respond with. they always end up thinking i'm a native speaker or a chinese person who's in america 😅
or im getting messed with and i shouldn't respond to these texts either way i've had some funny conversations that usually stop as soon as they find out i'm american using a translator.
I'm using deepl long time ago and the quality is better than bing and google's translator.
Agreed! It doesn’t cover as many languages as the others but tends to be more accurate.
I'm very excited about the fact that translation is becoming local to devices. Amethyst uses ML Kit to do this. Some of the translations are pretty funny/bad but as the models improve and more and more devices have ML accelerators, I don't think this will remain an issue for long.
I also have an libretranslate endpoint that can be used without API keys at translate.thesamecat.io
It isn't the best, but you can configure Damus to use it, and I do on iOS.
Nice! Thanks for running it. Mine is setup as libretranslate compatible as well so works in Damus. I do think running models locally is ideal but the centralized services have a big edge in accuracy atm.
Very much so. Larger model and more power. But it is getting better for sure. Google Translate allows you to download individual languages for offline use (on Pixel at least, I imagine is true on other androids)
I would love to see them open licence those models. Google bad, and all that, but we occasionally get lucky with them.
Can translations be cached on your provider end, or are their terms of service issues?
Obviously multiple language pairs per event, however I’d expect that it’s biased to a few major pairs.
I’m caching for myself today. I’m not a lawyer and haven’t reviewed the terms in detail but can’t imagine someone could tell me I can’t store strings of text. Perhaps dicier if I started selling access but I’d take a deeper dive first.
Yeah. I’ve just seen some really stupid terms in the past around some former translation services saying you cannot cache results or whatever. Unsure how legit it is - impossible to really enforce anyway.
Literally anything that’s user facing needs caching today as it scales to have decent UX anyway.
Exactly! I wouldn’t be surprised if one of the providers has hidden something like this on page 127 of their TOS but I’m skeptical that it’s enforceable.
7.5 DeepL does not assume any copyrights to the translations made by Customer using the Products. In the event that the translations made by Customer using the Products are deemed to be protected under copyright laws to the benefit of DeepL, DeepL grants to Customer, upon creation of such translations, all exclusive, transferable, sublicensable, worldwide perpetual rights to use the translations without limitation and for any existing or future types of use, including without limitation the right to modify the translations and to create derivative works.
Probably not a silver bullet but encouraging.
I would use it
Nice! Might just be you and me then 😂
Gonna test it myself for a couple days and make sure I didn’t miss anything obvious then I’ll set it free.
That would be great to participate anything for me that AI and Nostr related.
