According to the video you don't seem to automatically translate everything - thankfully! - but should you plan this in the future, please, please give your users an opt-out or a setting.
Just minutes ago I wrestled again with Youtube which doesn't respect my language preferences and tries to autotranslate not just the titles but also dubs videos in a shitty AI-translation, which I don't want. Youtube doesn't really seem to give you settings for this.
Machine translation may have gotten better, but it is still not right if you speak both languages. It get's the information across but always feels like the uncanny valley, it doesn't feel right. So my preference still is to never have machine translation between languages I already understand.
Automatically translating everything can be expensive (in terms of API cost, waiting period for the user). So I didn't want to keep it enabled for all posts.
I found that LLMs were much better quality at translation than Google Translate, especially long form text.
I created app.698expat.com because I was so frustrated at how bad google translate can be when translating user generated content (like facebook posts) or messages that require additional context.
For example, in Vietnamese, you need to know the age of the person relative to the speaker to properly address them. If you translate "Hello", Google translate says "Xin Chao" (which is the proper, but bland translation). A more conversational translation would be "(da) chao (anh|chi|co|con|ong|..)", but in order to know which last word to use, the translator needs more context.
example:
me: I am speaking to an older woman. how do I say hello to her in vietnamese?
gpt: When speaking to an older woman in Vietnamese, a polite and respectful way to say hello is: "Chào cô" – if she’s around your parents’ age (like an aunt).
I initially explored couple of LLM/NLP based translation models that could run on CPU hardware. But I couldn't get it to work with my Rust backend app (I'm noob to Rust and vibe coding doesn't work). Also need to consider the hardware compute required to run these models, as it can be expensive.
Are you using the Google Cloud Translate API or something? AFAIK google translate is a paid API, curious how you're doing it in a free / open source app. BYOK?
I found this Rust crate rust-translate (https://github.com/Mintype/rust-translate) that seems to be a wrapper around Google Translate. I guess the Google API gives some free tier for basic usage so it works? Idk, you might want to check try it out.
According to the video you don't seem to automatically translate everything - thankfully! - but should you plan this in the future, please, please give your users an opt-out or a setting.
Just minutes ago I wrestled again with Youtube which doesn't respect my language preferences and tries to autotranslate not just the titles but also dubs videos in a shitty AI-translation, which I don't want. Youtube doesn't really seem to give you settings for this.
Machine translation may have gotten better, but it is still not right if you speak both languages. It get's the information across but always feels like the uncanny valley, it doesn't feel right. So my preference still is to never have machine translation between languages I already understand.
Automatically translating everything can be expensive (in terms of API cost, waiting period for the user). So I didn't want to keep it enabled for all posts.
I also did a thing in my bookmarking software.
I added advanced button, for any link, which shows menu where you can navigate to Google translate, internet archive, schema validation, whois pages.
With the menu I can check links, and navigate easily
https://github.com/rumca-js/Django-link-archive/blob/main/RE...
I found that LLMs were much better quality at translation than Google Translate, especially long form text.
I created app.698expat.com because I was so frustrated at how bad google translate can be when translating user generated content (like facebook posts) or messages that require additional context.
For example, in Vietnamese, you need to know the age of the person relative to the speaker to properly address them. If you translate "Hello", Google translate says "Xin Chao" (which is the proper, but bland translation). A more conversational translation would be "(da) chao (anh|chi|co|con|ong|..)", but in order to know which last word to use, the translator needs more context.
example:
me: I am speaking to an older woman. how do I say hello to her in vietnamese?
gpt: When speaking to an older woman in Vietnamese, a polite and respectful way to say hello is: "Chào cô" – if she’s around your parents’ age (like an aunt).
I initially explored couple of LLM/NLP based translation models that could run on CPU hardware. But I couldn't get it to work with my Rust backend app (I'm noob to Rust and vibe coding doesn't work). Also need to consider the hardware compute required to run these models, as it can be expensive.
Are you using the Google Cloud Translate API or something? AFAIK google translate is a paid API, curious how you're doing it in a free / open source app. BYOK?
I found this Rust crate rust-translate (https://github.com/Mintype/rust-translate) that seems to be a wrapper around Google Translate. I guess the Google API gives some free tier for basic usage so it works? Idk, you might want to check try it out.