Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What Google Translate can tell us about vibecoding

ingrids.space

Post reply on HN
Print view View original

131–140 of 175 posts

Re: What Google Translate can tell us about vibecoding

#131

Post by Workaccount2 »

yoden wrote: post #124

Machine translation is a great example. It's also where I expect AI coding assistants to land. A useful tool, but not some magical thing that is going to completely replace actual professionals. We're at least one more drastic change away from that, and there's no guarantee anyone will find it any time soon. So there's not much sense in worrying about it. A very similar story has been happening in radiology for the p…

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture.

We don't know yet how a modern transformer trained on radiology would perform, but it almost certainly would be dramatically better.

Re: What Google Translate can tell us about vibecoding

#132

Post by the_af »

falcor84 wrote: post #125
the_af wrote: post #118

Earlier quoted context omitted.

Same. I don't understand why people want (them and their kids) to be isolated from other cultures. If you're watching a movie set in Japan, India or China, have it be about their culture. If there's something you or your kids don't understand, make an effort to learn about it (and the green peas thing seems trivial to understand). Netflix also does something absurd with their subtitles. I was watching "The Empress" (…

Could that Netflix subtitle thing have been a one-off error? I don't think I've ever encountered such a mismatch before. It did remind me of watching "The Beast" (La Bête)[0] in the original French with subtitles and I was then surprised when I saw the subtitles say "[In English]" and I was, "Oh, damn, the characters did actually switch to English". [0] https://en.wikipedia.org/wiki/The_Beast_(2023_film)

It's consistent through The Empress, I just gave an example. But maybe it's a decision specifically for this show?

For example, when Elisabeth is practicing several languages, each is subtitled "[in $language]", but when she switches back to German the subtitles readily explain... "[in English]". This confused the hell out of me!

Re: What Google Translate can tell us about vibecoding

#133

Post by tanvach »

Alex-Programs wrote: post #102
tkgally wrote: post #5

> ... a translators’ and interpreters’ work is mostly about ensuring context, navigating ambiguity, and handling cultural sensitivity. This is what Google Translate cannot currently do. Google Translate can't, but LLMs given enough context can. I've been testing and experimenting with LLMs extensively for translation between Japanese and English for more than two years, and, when properly prompted, they are really go…

I've ended up doing a lot of research into LLM translation, because my language learning tool ( https://nuenki.app ) uses it a lot. I built something kinda similar, and made it open source. It picks the top x models based on my research, translates with them, then has a final judge model critique, compare, and synthesise a combined best translation. You can try it at https://nuenki.app/translator if you're interested…

Very neat, love how there’s a formality level selection! Google translate has such bad tendencies to use very formal language (at least when translating into Thai) that it’s almost useless in real life. Some English to Thai examples I tried so far have been quite natural.

Re: What Google Translate can tell us about vibecoding

#134

Post by p3rls »

I just posted that Google is translating words that should be translated like "bastard" in foreign pop music (in this case, a member of the most popular boy band in the world, BTS) as racial slurs like the n-word. This is pretty much the worst case error scenario for a translation service.

https://kpopping.com/news/2025/Jun/17/Google-Translation-of-...

It's amazing more people aren't talking about this stuff, how incompetent do you have to be to allow racial slurs in translation, especially with all the weights towards being pro-diversity etc that already exist?

Just more enshittification from ol Sundar and the crew

Re: What Google Translate can tell us about vibecoding

#135

Post by tkgally »

jedberg wrote: post #130
tkgally wrote: post #5

> ... a translators’ and interpreters’ work is mostly about ensuring context, navigating ambiguity, and handling cultural sensitivity. This is what Google Translate cannot currently do. Google Translate can't, but LLMs given enough context can. I've been testing and experimenting with LLMs extensively for translation between Japanese and English for more than two years, and, when properly prompted, they are really go…

> You give it a text to be translated ... and then sends that draft back to the three models for several rounds of revision, checking, and polishing. Interesting. Curious if you modeled the cost of that single translation with the multiple LLM calls and how that compares to a human.

I had Claude Code write a module that monitored the incoming and outcoming token counts and display the accumulated costs. A Japanese-to-English translation that yielded about a thousand words in English cost around US$0.40.

I didn't double-check the module's arithmetic, but it seems to have been in the ballpark, as my total API costs for OpenAI, Anthropic, and Google yesterday--when I was testing this system repeatedly--came to about eight dollars.

A human translator would charge many, many times more.

Re: What Google Translate can tell us about vibecoding

#136

Post by TechDebtDevin »

Workaccount2 wrote: post #131
yoden wrote: post #124

Machine translation is a great example. It's also where I expect AI coding assistants to land. A useful tool, but not some magical thing that is going to completely replace actual professionals. We're at least one more drastic change away from that, and there's no guarantee anyone will find it any time soon. So there's not much sense in worrying about it. A very similar story has been happening in radiology for the p…

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture. We don't know yet how a modern…

That wouldnt be that much different than current CNN/labeling methods used on medical imaging. Last time I got a ct scan the paperwork had the workstation specs and the models/nueral network techniques used.

Re: What Google Translate can tell us about vibecoding

#137

Post by tiagod »

resoluteteeth wrote: post #116
AndroTux wrote: post #86

Earlier quoted context omitted.

That’s absolutely not the point of this article. The point was that people said, once Google Translate was introduced, that translators would lose their jobs. Just like people say the same thing about developers with LLMs nowadays. The point is: they didn’t, and they won’t. DeepL is not part of that point. Yes, maybe eventually, developers will lose their jobs to something that is an evolution of LLMs. But that’s not…

> The point was that people said, once Google Translate was introduced, that translators would lose their jobs. Just like people say the same thing about developers with LLMs nowadays. The point is: they didn’t, and they won’t. Translators are losing their jobs now though. Google translate wasn't very good for Japanese so a lot of people assumed that machine translation would never be a threat, but deepl was better t…

I highly doubt LLMs will do a good job translating literature anytime soon.

Re: What Google Translate can tell us about vibecoding

#138

Post by abrugsch »

the_af wrote: post #118
falcor84 wrote: post #115

Earlier quoted context omitted.

Is that beans example a real thing? If so, I would hate for my kids to be subjected to that. The best thing about watching films from another country is that you're exposed to the culture of that foreign place and learn about how it's different from yours - I don't see why we'd try to localize away the human experience as if these differences don't exist.

Same. I don't understand why people want (them and their kids) to be isolated from other cultures. If you're watching a movie set in Japan, India or China, have it be about their culture. If there's something you or your kids don't understand, make an effort to learn about it (and the green peas thing seems trivial to understand). Netflix also does something absurd with their subtitles. I was watching "The Empress" (…

Just saying "Pixar movies" was probably not a great example. They can be deliberately location ambiguous (Monsters Inc., Toy Story - though it's clearly _somewhere_ in America, The Incredibles - a generic "metropolis"/50's futuristic city, lightyear, Elemental) or very specifically somewhere (cars - mashup of Route 66 towns, Finding Nemo - Sydney when on land, Ratatouille - Paris, etc...)

It makes sense to "translate" locale cultural indicators in say Wall·E which was very location agnostic but not so much for say Turning red which is very culturally specific.

Re: What Google Translate can tell us about vibecoding

#139

Post by demosthanos »

Workaccount2 wrote: post #131
yoden wrote: post #124

Machine translation is a great example. It's also where I expect AI coding assistants to land. A useful tool, but not some magical thing that is going to completely replace actual professionals. We're at least one more drastic change away from that, and there's no guarantee anyone will find it any time soon. So there's not much sense in worrying about it. A very similar story has been happening in radiology for the p…

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture. We don't know yet how a modern…

> We don't know yet how a modern transformer trained on radiology would perform, but it almost certainly would be dramatically better.

Why? Is there something about radiology that makes the transformer architecture appropriate?

My understanding has been that transformers are great for sequences of tokens, but from what little I know of radiology sequence-of-tokens seems unlikely to be a useful representation of the data.

Re: What Google Translate can tell us about vibecoding

#140

Post by dlisboa »

falcor84 wrote: post #115
philsnow wrote: post #6

The distinction between what people typically imagine a translator's job is and the reality reminds me of pixar movies being "localized" instead of just translated (green beans on a plate in the Japan release instead of broccoli because that's the food that Japanese kids don't like). Lacking cultural context while reading translated texts is what made studying history finally interesting to me.

Is that beans example a real thing? If so, I would hate for my kids to be subjected to that. The best thing about watching films from another country is that you're exposed to the culture of that foreign place and learn about how it's different from yours - I don't see why we'd try to localize away the human experience as if these differences don't exist.

Localization is more important than you might think.

My wife worked for a company that helped provide teaching content for schools throughout Brazil. They'd interview teachers all over the country and one of the complaints from teachers in isolated communities was that they had to use the same textbooks as other places in Brazil without any regard to their own situation.

They reported that many examples for starting math for kids featured things like "strawberries" or "apples", things the kids had never seen or maybe heard. So now they needed to abstract over what is a "fruit" and a "countable object" as well as whatever the example was trying to teach. Teachers reported less engagement and it was more work for them to adapt it to local relevance.

Try to teach kids about vegetables in the US midwest and use green beans and Bok Choy as an example, for instance. It doesn't make sense.

Post reply on HN