Smaller companies that can afford smaller salaries you mean?
I'm not sure that is at odds with what the post you were replying to said
https://cloud.google.com/translate/docs/advanced/translating...
DeepL also has a translation LLM, which they claim is 1.4-1.7x better than their classic model: https://www.deepl.com/en/blog/next-gen-language-model
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 relevant here.
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 to the point where a lot of translation moved to just cleaning up it's output and current state of the art llms as of the last six months are much better and can also be given context and other instructions to reduce the need for humans to clean up the output. When the dust settles translation as a job is probably going to be dead.
It's the equivalent of llms eliminating everything except a handful of system architect jobs at FAANG companies in terms of programming.
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 good. I say this as someone who worked for twenty years as a freelance translator of Japanese and who still does translation part-time.
Just yesterday, as it happens, I spent the day with Claude Code vibe-coding a multi-LLM system for translating between Japanese and English. You give it a text to be translated, and it asks you questions that it generates on the fly about the purpose of the translation and how you want it translated--literal or free, adapted to the target-language culture or not, with or without footnotes, etc. It then writes a prompt based on your answers, sends the text to models from OpenAI, Anthropic, and Google, creates a combined draft from the three translations, and then sends that draft back to the three models for several rounds of revision, checking, and polishing. I had time to run only a few tests on real texts before going to bed, but the results were really good--better than any model alone when I've tested them, much better than Google Translate, and as good as top-level professional human translation.
The situation is different with interpreting, especially in person. If that were how I made my living, I wouldn't be too worried yet. But for straight translation work where the translator's personality and individual identity aren't emphasized, it's becoming increasingly hard for humans to compete.
Disclaimer: I work for Soniox.
I live in Japan. Almost every day I find myself asking things like “what does X mean in this specific setting?” or “how do I tell Y to that specific person via this specific medium?”.
Much of this can be further automated via custom instructions, so that e.g. the LMM knows that text in a particular language should be automatically translated and explained.
Great ideas. I'll think about adding those features to the system in my next vibe-coding session.
What I automated in the MVP I vibe-coded yesterday could all be done alone by a human user with access to the LLMs, of course. The point of such an app would be to guide people who are not familiar with the issues and intricacies of translation so that they can get better translations for their purposes.
I have no intention to try to commercialize my app, as there would be no moat. Anyone who wanted to could feed this thread to Claude, ask it to write a starting prompt for Claude Code, and produce a similar system in probably less time than it took me.
The word "potatoes" in context of a specific 500-page book has little ambiguity. Same word but extracted out of the same book and fed to a translator(human or machine) in isolation would be much more ambiguous. You probably don't need the whole book, but the solution space do reduce as you give translators more out of the content or how it's used in the original as well as in other parts of translations.
It's similar to how GPS works. With one satellite, your location is precise as "on Earth, I guess". It gets more precise as you add more satellites that further and further reduce margins of errors.
It can be as simple as discuss one’s own religion
I also have a few heuristics (e.g. "I can't translate" in many different languages) to detect if it deviates from that.
It works pretty well.
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, and my data is at https://nuenki.app/blog
https://nuenki.app/blog/the_more_llms_think_the_worse_they_t...
Your results accord with my own (much less systematic) tests of the translation of short texts by reasoning models. The issue becomes more fuzzy with the translation of longer texts, where quality is more difficult to evaluate objectively. I'll drop you an email with some thoughts.
As a human translator, if I were starting to translate a text in the middle and I wanted my translation to flow naturally with what had been translated before, I would want both a summary of the previous content and notes about how to handle specific names and terms and maybe about the writing style as well. When I start working on the project again tomorrow, I’ll see if Claude Code can come up with a solution along those lines.
The only thing that actually worked was knowing the target language and sitting down with multiple LLMs, going through the translation one sentence at a time with a translation memory tool wired in.
The LLMs are good, but they make lot of strange mistakes a human never would. Weird grammatical adherence to English structures, false friend mistakes that no one bilingual would make, and so on. Bizarrely many of these would not be caught between LLMs -- sometimes I would get _increasingly_ unnatural outputs instead of more natural outputs.
This is not just for English to Asian languages, even English to German or French... I shipped something to a German editor and he rewrote 50% of the lines.
LLMs are good editors and suggestors for alternatives, but I've found that if you can't actually read your target language to some degree, you're lost in the woods.
I have been astounded at the sophistication of LLM translation, and haven't encountered a single false-friend example ever. Maybe it depends a lot on which models you're using? Or it thinks you're trying to have a conversation that code-switches mid-sentence, which is a thing LLM's can do if you want?
Interesting. Curious if you modeled the cost of that single translation with the multiple LLM calls and how that compares to a human.
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.
How is this part done? How are they chosen/combined to give the best results? Any info would be appreciated as I've seen this sort of thing mentioned before, but details have been scant!
https://www.gally.net/temp/20250619translationprogramtrace/i...
The Anthropic servers seem to have been overloaded, but if you read through the above it should give you an idea of how the program is currently doing that synthesis. If you want the code as it stands now (a single HTML file), feel free to email me. My website is linked from my profile page.
RIP global power consumption
Lacking cultural context while reading translated texts is what made studying history finally interesting to me.
First, the Pixar thing was green pepper, not green beans: https://www.businessinsider.com/why-inside-out-has-different...
Second, the Pixar one is not "mere" translation; it is full localization because they changed the visual to match the "textual" change.
The Pokemon one is where the change was limited to the "text". The translator's heart might have been in the right place (it would depend on how integral to the story it is that the item is onigiri) but didn't have the authority to make the full breadth of changes needed for such adaptation to be successful.
4Kids was very well known to visually change the japanese shows they imported if they thought it was worth it, mostly in the context of censorship. For example, all guns and cigarettes where removed from One Piece, turned into toy guns and lollipops instead.
The most infamous example, however, has got to be Yu-Gi-Oh!. Yu-Gi-Oh started as a horror-ish manga about a trickster god forcing people to play assorted games and cursing their souls when they inevitably failed to defeat him. The game-of-the-week format eventually solidified into the characters playing one single game, Duel Monsters (the Yu-Gi-Oh! TCG itself in the real world), and the horror-ish aspects faded away, although they still remained part of the show's aesthetic, based around Egyptian human sacrifices and oddly-card-game-obsessed ancient cults.
When the manga was adapted to the screen, it started directly with a softer tone[1], especially because the show was to be a vehicle for selling cards in the real world, not dissimilarly to Pokemon and MANY other anime from the era.
Nothing that happens in the show is particularly crude or shocking, it had that kind of soft edginess that fit well with its intended target audience (early teens). I imagine watching Bambi had to be much more traumatizing than anything in the show.
But that was still not enough for 4Kids, which had a pretty aggressive policy of no violence or death. Kind of problematic when the show's main shtick was "Comically evil villain puts our heroes in a contraption that will kill them if they don't win." (You can imagine the frequency these traps actually triggered neared zero).
To solve this, 4Kids invented the Shadow Realm. The show, thanks to its occultist theming, already had examples of people being cursed, or their souls being banished or captured. 4Kids solidified these vague elements into the shadow realm as a censorship scape-goat. Any reference to death was replaced with the shadow realm. Now, one might wonder why the censors thought that "hell-like dimension where your soul wanders aimlessly and/or gets tortured for eternity" was in any way less traumatizing than "you'll die", but I imagine it's because there was always the implication that people could be 'saved' from the shadow realm[2] by undoing the curse.
The Shadow Realm was a massive part of the western Yu-Gi-Oh mythos and even today it's a fairly common meme to say that somebody got "sent to the shadow realm", which makes it all funnier that it is not part of the original show.
A couple funny examples off the top of my head: - Yugi must win a match while his legs are shackled. Two circular saws, one for him and one for the enemy, are present in the arena. They near the two competitors as they lose Life Points, with the loser destined to have their legs cut off.
In the 4Kids adaptation, the saws are visually edited to be glowing blue, and it's stated they're made out of dark energy that will send anybody that touches it to the shadow realm.
- A group of our heroes fight a group of villains atop of a skyscraper with a glass roof. In the original version, the villains state that the roof has been boobytrapped so that the losing side will explode, plunging the losers to their death by splattening.
In the 4Kids version, the boobytrap remained, but the visuals were edited to add a dark mist under the glass, with the villains stating that there's a portal under the roof that will send anybody that touches it to the shadow realm. This is made funnier when the villains lose and they're shown to have had parachutes with them all along, and they are NOT edited out.
[1] Technically speaking, there was a previous adaptation that followed the manga more closely and got only one season, generally referred to as Season 0.
[2] It does eventually happen in the anime that the heroes go in an alternate dimension to save somebody's cursed soul. Obviously, this dimension was directly identified as the Shadow Realm in the localization.
Netflix also does something absurd with their subtitles. I was watching "The Empress" (which is set in the Austro-Hungarian Empire) with German language and English subtitles. I like listening to the real actors' voices, and learning the sounds and cadence of the language. So the characters speak in Italian for a while (subtitles say "[speaking in Italian]", and when they switch back to German the subtitles clarify.. "[in English]". The fuck, Netflix? Surely the viewer of this show understands they didn't speak in English in the Austro-Hungarian empire, so why write it's English? What the hell is Netflix even trying to achieve here? Seems robotic: "us = English speakers, therefore everyone's default must be English"?
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".
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!
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.
The localization of The Incredibles in Argentina was embarrassing, though someone must have thought it was a good idea. They used local voice actors popular at the time (though not so much today) with strong porteño (Buenos Aires') accents. They also referred to the streets with Argentinian names, e.g. "let's turn that corner of Corrientes Avenue!". The problem is that Corrientes Av is very typical of Buenos Aires, but nothing on screen looked anywhere close to it, so the whole thing was ridiculous and embarrassing, sort of like if the characters pretended they were in Tokyo and were Japanese.
What if they had gone the extra mile (maybe possible in the near future) and reskinned every character to look more Argentinian, and rethemed the city to look more like Buenos Aires, would I have been happier? Certainly not -- I want to see the original movie as intended, not some sanitized version designed to make me feel "at home" and not challenge me in the slightest.
(I watched the movie in English as well, mind you).
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.
We bought kids toys on Amazon and the fruits were strange. Not sure if they were Asian varieties or just made up.
> I feel confident in asserting that people who say this would not have hired a translator or learned Japanese in a world without Google Translate; they’d have either not gone to Japan at all, or gone anyway and been clueless foreigners as tourists are wont to do.
The correlation here would be something like: the people using AI to build apps previously would simply never have created an app, so it’s not affecting software development as a career as much as you first expect.
It would be like saying AI art won’t affect artists, because the people who would put in such little effort probably would never have commissioned anyone. Which may be a little true (at least in that it reduces the impact).
However, I don’t necessarily know if that’s true for software development. The ability to build software enabled huge business opportunities at very low costs. I think the key difference is this: the people who are now putting in such low effort into commissioning software maybe did hire software engineers before this, and that might throw off a lot of the numbers.
It's all cleaning up slop code. Always has been.
</tired old fart voice>
More optimistically, you can think of "user created code" as an attempt at a design document of sorts; they were trying to tell you (and the computer) what they wanted in "your language". And that dialog is the important thing.
We can never decide, we just like learning, and there is little real, impactful research into programming as a business.
In two decades we will still collectively say ”we are learning so much”, ignoring that fact.
The work got easier, so what we do got more complex.
Becoming obsolete is a fear of people who are not willing or able to learn arbitrary problem domains in a short amount of time. In that case learning to use a particular tool will only get you so far. The real skill is being able to learn quickly (enthusiasm helps).
Gas powered pogo sticks, shoe fitting X-ray, Radium flavored chocolates, Apollo LLTV, table saws, Flex Seal for joining two halves of boats together, exorbitantly parallelized x86 CPU, rackable Mac Pro with M1 SoC, image generation AI, etc.
Tools can be useless, or be even dangerous.
How processors work, cache and memory work, how the browser works, data structure and algorithms, even design patterns are all important foundationaly knowledge. How to tell an AI to shit out some code or answer a question definitely isn't.
I know enough Japanese to talk like a small child, make halting small talk in a taxi, and understand a dining menu / restaurant signage broadly. I also have been enough times to understand context where literal translation to English fails to convey the actual message.. for example in cases where they want to say no to a customer but can't literally say no.
I have found Google Translate to be similarly magical and dumb for 15 years of traveling to Japan without any huge improvements other than speed. The visual real-time image OCR stuff was an app they purchased (Magic Lens?) that I had previously used.
So who knows, maybe LLM coding stays in a similar pretty-good-never-perfect state for a decade.
I think this is definitely a possibility, but I think the technology is still WAY too early to know that if the "second AI winter" the author references never comes, that we still wouldn't discover tons of other use cases that would change a lot.
Word Lens, by Quest Visual
Meanwhile, the point of software development is not to write code. It's to get a working application that accomplishes a task. If this can be done, even at low quality, without hiring as many people, there is no more value to the human. In HN terms, there is no moat.
It's the difference between the transition from painting to photography and the transition from elevator operators to pushbuttons.
I'm not sure how seriously people take the threat of non-coding vibe-coders. Maybe they should! The most important and popular programming environment in the world is the spreadsheet. Before spreadsheets, everything that is today a spreadsheet was a program some programmer had to write.
The normal conversation is that productivity growth has slowed and the divide has increased, not that more productivity creates lower outcomes in real terms.
https://www.bls.gov/productivity/images/labor-compensation-l...
Data is collected through the National Compensation Survey: https://www.bls.gov/respondents/ncs/
If AI coding improves productivity, it might move us closer to having 2X as much work as we can possibly do instead of 3X.
Nowadays we already have bullshit jobs that keep academics employed. Retraining takes several years.
With "AI" the danger is theoretically limited because it creates more bureaucracy and reduces productivity. The problem is that it is used as an excuse for layoffs.
They don't know the office politics, or go on coffee breaks with the team - humans still have more context and access. We still need people to manage the goals and risks, to constrain the AI in order to make it useful, and to navigate the physical and social world in their place.
The AI slop will make it harder for the small guys without marketing budget (some lucky few will still make it though). It will slowly kill the app ecosystem, untill all we reluctantly trust is FANG. The app pricing reflects it.
That is incorrect, sir.
First, because many problems were designed to fit into spreadsheets (human systems designed around a convenient tool). It is much more likely that several spreadsheets were _paper_ before, not custom written programs. For a lot of cases, that paper work was adapted directly to spreadsheets, no one did a custom program intermediate.
Second, because many problems we have today could be solved by simple spreadsheets, but they often aren't. Instead, people choose to hire developers instead, for a variety of reasons.
I say spreadsheet software replace paper.
That's the disagreement. You have intuition, I have sources:
https://en.wikipedia.org/wiki/Spreadsheet#Paper_spreadsheets
https://en.wikipedia.org/wiki/Spreadsheet#Electronic_spreads...
The other is about an "AI" website generator, spamming every video at the start.
I wonder what kind of honest efforts would require that kind of marketing.
I don't think the original point or your interpretation is correct.
AI will not cause a loss of software development jobs. There will still be a demand for human developers to create software. The idea that non-technical managers and executives will do so with AI tools is as delusional as it was when BASIC, COBOL, SQL, NoCode, etc. were introduced.
AI will affect the industry in two ways, though.
First, by lowering the skill requirements to create software it creates a flood of vibe coders competing for junior-level positions. This dilutes the market value of competent programmers, and makes entering the software industry much more difficult.
A related issue is that vibe coders will never become programmers. They will have the ability to create and test software, which will improve as and if AI tools continue to improve, but they will never learn the skills to debug, troubleshoot, and fix issues by actually programming. This likely won't matter to them or anyone else, however, but it's good to keep in mind that theirs is a separate profession from programming.
Secondly, it floods the software market with shoddy software full of bugs and security issues. The quality average will go down causing frustration for users, and security holes will be exploited increasing the frequency of data leaks, privacy violations, and unquantifiable losses for companies. All this will likely lead to a rejection of AI and vibe coding, and an industry crash not unlike the video game one in 1983 or the dot-com one in 2000. This will happen at the bottom of the Trough of Disillusionment phase of the hype cycle.
This could play out differently if the AI tools reach a level of competence that exceeds human senior software engineers, and have super-human capabilities to troubleshoot, fix, and write bug-free software. In that case we would reach a state where AI could be self-improving, and the demand for human engineers would go down. But I'm highly skeptical that the current architecture of AI tools will be able to get us there.
Google Translate is doing a bad job.
The Chrome translate function regularly detects Traditional Chinese as Japanese. While many characters are shared, detecting the latter is trivial by comparing unicode code points - Chinese has no kana. The function used to detect this correctly, but it has regressed.
Most irritatingly of all, it doesn't even let you correct its mistakes: as is the rule for all kinds of modern software, the machine thinks it knows best.
The syntaxes aren't just different but generally backwards, and, it's just my hunch but, they sometimes sound like they are confused about which modifies word which.
While professionals still produce much better quality translations, the demand for everything but the most sensitive work is nearly gone. Would you recommend your offspring get into the industry?
Now a professional service only needs a reviewer to edit what the LLM produces. You can even ask the LLM to translate in certain tones, dialects, etc. and they do it very well.
* Idioms (The article mentions in passing that this isn't so much a difficulty in Norwegian->English, but of course idioms usually don't translate as sentences)
* Cultural references (From arts, history, cuisine, etc. You don't necessarily substitute, but you might have to hint if it has relevant connotations that would be missed.)
* Cultural values (What does "freedom" mean to this one nation, or "passion" to this other, or "resilience" to another, and does that influence translation)
* Matching actor in dubbing (Sometimes the translation you'd use for a line of a dialogue in a book doesn't fit the duration and speaking movements of an actor in a movie, so the translator changes the language to fit better.)
* Artful prose. (AFAICT, LLMs really can't touch this, unless they're directly plagiarizing the right artful bit)
LLM’s will tell you idioms, slang, and the point behind it
You can take a screenshot of telegram channels for both sides of a war conflict and get all context in a minute
In classic HN fashion I’m sure I missed the point, ok translators are still in demand got it.
Google Translate has been leapfrogged by the same thing that allows for “vibecoding”
> At the dinner table a Norwegian is likely to say something like “Jeg vil ha potetene” (literally “I will have the potatoes”, which sounds presumptuous and haughty in English). Google Translate just gives the blunt direct translation.
"To will" means "it is my will", i.e. to want to, which became the future tense in English. In Norwegian is still means "want", and Google Translate indeed translates it as "I want the potatoes." If you translate the rising (pleading) intonation on "potatoes", you then have an unwritten "please?", i.e. "I want the potatoes?...", which is passable English.
Most businesses think AI code is "good enough", and that machine translation is "good enough", which tanks the entire industry because there is now more supply than demand. He says there are still plenty of translator jobs, but then justifies it as because "it’s inadvisable to subsititute (sic) Google Translate for an interpreter at a court hearing." Meaning, the thing taking away all the tech jobs temporarily (unchecked mass-migration) is the same thing keeping keeping him employed temporarily.
I find “I will have the potatoes” to be perfectly fine English and not haughty in the slightest. Is this a difference between British English and American English?
When asking someone to pass them to you, just imagine them turning to you, looking you in the eye, and asserting "I will have the potatoes" like it's some kind of ultimatum. Yes, that is strange.
I stopped saying, stuff "I would like a latte today" or more Midwestern (could I get a latte today etc) in singapore because people would just get confused. Same with being too polite when recieving things. There's ways to be polite but it usually involves less words because anything else confuses people.
To me (non-American) the above sounds like sarcasm, not politeness. Adding "today" and/or "actually" could mean you've had it with their delays.
I like to joke that Americans always seem to find ways to get offended by innocuous things, but in this case the joke is on me.
To me (Canadian, not American) "Could I get a latte today actually" sounded something like "Normally I get something other than a latter but actually today I would like a latte instead"
Not rude at all, but kind of assumes some context
To me it seems (without any context) that it might mean that you changed your mind about what day you wanted it. This does not seem to make sense in many contexts, though.
Having grown up in the UK and living in Australia, I do not find Americans polite. To me, politeness is "please", "thank you", "may I have", etc, whereas "I would like a latte today" reads to me as a demand. In context it's fine (it stands out a bit but not in a problematic way), it's not particularly rude, but in general just stating your desires is not considered polite in my experience in UK/AU.
There are some other parts of American English that may be considered polite, I notice a lot of US folks addressing me as "sir" for example, but this sort of thing comes off as insincere or even sarcastic/rude.
I know this is how people communicate so they don't really bother me, I'm used to them and accept that different people have different expectations. I also understand that Americans might believe they are being polite and consider me to be rude, but I think this is why blanket statements like "Americans are extremely polite" are just missing so much cultural nuance.
I've never ordered a latte "today".
A simple statement of "I'd like a latte" fits fine in our regional vernacular.
I think that "I'd like a latte today" would appear a weird bit superfluous of specificity, unless it comes from a regular and familiar customer who might normally order something different and/or random.
I mean: "Today"? As opposed to later this week or something? It's implicit that the latte is wanted as soon as they can get around to it. (Unless yesterday's order was something different, and then adding "today" may make sense.)
---
But English is fucked up, and I'm getting old, so I've spent a lot of time observing (and sometimes producing) slight miscommunication.
In terms of things like ordering food or drink at a counter, it still works fine as long as nobody gets grumpy, and the desired item is produced.
I'm reminded of a local festival I went to as a little kid, with my parents, sometime. I was getting hungry and they were asking me what I wanted to eat. There were a lot of pop-up food vendor there, mostly with tables and stuff instead of the now-ubiquitous "food truck."
In the corner was a gyro stand with amazing-looking racks of spinny-meat. I wanted to try whatever that was.
The big banner said "GYROS" and we got in line.
Discussions were had between my parents about this "GYROS" concept, which they'd never seen before either. Was it a singular, or a plural? How many "GYROS" might a boy reasonable want? How was it pronounced? It looks like "gyro" as in "gyrocopter" but it seemed to them that it must be Greek, so they went through some variations on pronunciation as the line moved forward.
As we got closer, some of these questions were answered: The sign definitely referred to a plural offering, and seeing people leave it became clear that [unlike things like chorros or tacos or donuts] one was plenty.
And when we got to the front, the conversation went like this:
Parents: "Our son wants one of these... but we're not sure how to say it. Jye-roe? Hee-roe?"
"They're just gyros," he replied to them to them dismissively in a thick Greek accent, and motioned for his staff to produce 1 gyro.
And then the man looked at me, with his skin sweaty on that hot sunny day and a thick gold chain around his neck, and said to me in his very best and most careful English something to me that I can never forget. "I call them gyros. But listen to me, you can call them whatever you want. Jye-roe? Hee-roe? Yee-roe? Sandwich? Whatever you say, and however you say it: If you get what you expect, then you said it right. Alright?"
My trepidatious nods made sure that he was understood, and the awesome fucking sandwich-taco I had that day changed my entire perspective on food -- and ordering food -- forever.
So, sure: Ordering "one latte today"?
It sounds weird to me, but if it is understood and you get what you want, then it works. Correctness? Politeness? Whatever. Despite seeming weird: It works.
(Up next: There's a lot of ways to mispronounce approximately everything related to ordering pho, and none of them are wrong.)
That's news to me! In my part of the world I call even the janitorial staff "Sir". I'm not aware of anyone ever thinking that was rude.
When an American says "sir" to me, I know I'm not in any of those positions, so it comes across as parody of those – highlighting the fact that I am not actually in the position of power, or that I am not a highly valued customer. In that way it's sarcastic, and it can imply that I'm actually imposing on the other person and taking that position of power, which as I'm not, therefore comes across as passive aggressive in a way.
If an American asked me to do something and I said "yes Mr President", arguably referring to the highest status in the culture, I expect it would be taken as sarcastic or rude, not polite, because the person is not in fact the president. That's sort of how I see it, but maybe this analogy doesn't work.
This makes sense since the context is translation for tourism.
Otherwise, the normal, casual way would be "kan du sende potetene?" i.e. "could you pass the potatoes?", lit. "can you send the potatoes?"
(This assumes it wasn't physically possible to simply reach across people to grab it yourself with what's known as "the Norwegian arm")
It's similar to vibe coding where the user truly doesn't know how to create the same thing themselves: You end up with output that you have no way of knowing it's correct, and you just have to blindly trust it and hope for the best. And that just doesn't work for many situations. So you still need expertise anyway (either yours or someone else's)
The goal of software is not to compile and run, it's to solve a problem for the user when running. If you don't know enough about the problem to be able to assess the output of the program then you can't trust that it's actually doing the right thing.
I've seen code generated by LLMs that runs, does something, and that the something would be plausible to a layperson, but that I know from domain knowledge and knowing the gotchas involved in programming that it's not actually correct.
I think the comparison with translation is a very apt one.
So this test fails many times, and even when not, the translation is still usually not good. However, when you don’t care about nuances, it’s still fine usually. And also translation seems to be better if the text is larger. Translating a single word does almost always fail. Translating a whole article is usually fine. But even there it matters what you translate. Translating the Austrian subreddit fails quite often. Sometimes completely for whole large posts. Translating news is better.
e.g. "Set up database as: [read replica]" and "Database setup complete: [a replica was loaded.]" may translate into a same string in some language.
Only because you think the machine will be incorrect. If I have a large multiplication to do it's much easier to trust the calculator than a mathematician, and that's all due to my perception of the calculator accuracy.
No, the one side is saying that all their code is written by LLMs already and that's why they think that. In fact, I would say the other side is the former ("it works for throwaway code but that's it") and that no-one is flat-out rejecting it.
Fantastic closing paragraph! Loved the article
https://www.osnews.com/story/142469/that-time-ai-translation...
A very similar story has been happening in radiology for the past decade or so. Tech folks think that small scale examples of super accurate AIs mean that radiologists will no longer be needed, but in practice the demand for imaging has grown while people have been scared to join the field. The efficiencies from AI haven't been enough to bridge the gap, resulting in a radiologist _shortage_.
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.
I can only imagine what people picture when they think about AI and radiology, but I can certainly imagine that it goes beyond that. What if you can feed a transformer with literature on how the body works and diseases, so that it can _analyse_ (not just _classify_) scans, applying some degree of, let's call it, creativity?
That second thing, if technically feasible, confabulations and all, has the potential to replace radiologists, maybe, if you're optimistic. Simple image classification, probably not, but it sounds like a great help to make sure they don't miss anything, can prioritise what to look at, and stuff like that.
Then it would make up plausible-sounding nonsense, just like it does in all other applications, but it would be particularly dangerous in this one.
> 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.
In the last few years, using only my primitive knowledge of Japanese (and Chinese -- which helps a lot with reading/writing), I have been able to fill out complex legal and tax documents using my knowledge of Japanese and the help of Google Translate. When I walk into a gov't office as the only non-Asian person, I still get a double take, but then they review my slightly-less-than-perfect submission, and proceed without issue. (Hat tip to all of the Japanese civil servants who have diligently served me over the years.)
Hot take: Except for contracts and other legal documents, "actual professionals" (translators) is a dead career at this point.
DeepL is a step up, and modern LLMs are even better. There's some data here[0], if you're curious - DeepL is beaten by 24B models, and dramatically beaten by Sonnet / Opus / https://nuenki.app/translator .
[0] https://nuenki.app/blog/claude_4_is_good_at_translation_but_... - my own blog
Quote from article:
> it turns out the number of available job opportunities for translators and interpreters has actually been increasing.
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
- Google Translate has always been garbage for professional translation work, which is why human translators have been needed.
- LLMs can write in a way that sounds native (at least in English), which is something that ML translation software was bad at. They can also understand context to some degree and can adopt tones. This is a huge leap forward that makes it suitable for a large majority of translation work _between common language pairs_ (Hindi to Thai is probably safe).
- The vast majority of translation work is commercial in nature, and most cases, "decent" is good enough. Cost is prioritized over occasional mistakes. Prices have been falling for years already. Many agencies now pay peanuts because they expect you to use AI and "just" do some light proofreading. Except that often the "light proofreading" was actually "heavy editing" that often took more work than translating from scratch. It's not worth it.
- Do not lump translators and interpreters together when discussing this subject. Interpreters aren't going away and their job is much more difficult to replace with AI (for a variety of reasons). What they do is very different than translators (some do both of course).
- There are some exceptions to "decent is good enough", where: 1) context/localization is critical (a major advertising campaign), 2) accuracy of the message is critical (a government pronouncement), 3) linguistic excellence is critical (a novel). This may be the most visible parts, but also a very small fraction of the overall market.
My conclusion is that translators won't disappear altogether but their ranks will shrink considerably.
Absolutely! AI might get better at context, but there's still no substitute for a seasoned dev who’s had to untangle a 2 a.m. deployment gone wrong. As for vibe coders — hey, good vibes are great, but they won’t refactor a legacy monolith on their own.
dr_dshiv•7mo ago
I want to believe there will be even more translators in the future. I really want to believe it.
alganet•7mo ago
Can you give us an example of a typical translation question and the "good prompting" required to make the LLM consider tone?
simonw•7mo ago
alganet•7mo ago
It includes a lot of steps and constant human evaluation between them, which implies that decisions about tone are ultimately made by whoever is prompting the LLM, not the LLMs themselves.
> "If they are generally in the style I want..."
> "choosing the sentences and paragraphs I like most from each..."
> "I also make my own adjustments to the translation as I see fit..."
> "I don’t adopt most of the LLM’s suggestions..."
> "I check it paragraph by paragraph..."
It seems like a great workflow to speed up the work of an already experienced translator, but far from being usable by a layman due to the several steps requiring specialized human supervision.
Consider the scenario presented by the blog post regarding bluntness/politeness and cultural sensitivities. Would anyone be able to use this workflow without knowing that beforehand? If you think about it, it could make the tone even worse.
nottorp•7mo ago
Just like programming. And anything else "AI" assisted.
Helps experts type less.
alganet•7mo ago
nottorp•7mo ago
alganet•7mo ago
I worked with Java for a couple of years. Lots of boilerplate required. The IDE handled them just fine. I never felt I needed something smarter in order to write verbose stuff.
My style is completely different though. Whenever I can, I will choose languages and technologies that don't require boilerplate.
https://wiki.c2.com/?BoilerPlateCode
However, this is deviating from the subject (look at the original post again). We're not talking about dumb, automated boilerplate code. The discussion is much deeper, drawing parallels with tone in translations. We're obviously talking about non-boilerplate.