The ending is a really powerful point. Most people apparently agree on two things:

1. AI is a great boon for all tasks and specialties we don’t have the skills to do ourselves. Understandable, since (A) we’re ill equipped to see the flaws in its output because it isn’t our area of expertise, and (B) it often can unlock great gains because if we trust it, we then don’t have to pay and wait for humans to do that thing.

2. AI is a terrible replacement for me - my skills are at such a high level that it’s almost theoretical that it’ll ever be good enough to replace me for 90% of what I get paid to do. It’s a tool at best.

This is why I use AI for all my medical questions and doctors use AI to write software, and we both smirk at the quality the other person is getting from it.

Slight tangent into translations:

I read two translations of the book "The Master and Margarita". My first read was so boring I couldn't help but stop reading before the end of the first chapter. I can't find the copy and the name of the person who translated it, but this one had all the Russian nicknames translated. It kept talking about a guy called homeless. I thought it was just a bad book and dismissed it for years. I couldn't understand what all the fuss was about with this book.

But then, I stumbled upon the translation by Diana Burgin and Katherine Tiernan O'Connor. Although I don't speak Russian, I think this is as good as it gets. They did a phenomenal job.

You can see the same effect with the mechanical translation of the book "We" by Yevgeny Zamyatin, where the government is called "United State" easily confused with the "United States". The translation that called it "One State" was so much better.

An honest to god article full of em dashes that's not because it was AI but because it was a human using them as a crutch to get around crafting sentences that flow naturally. Almost brings a tear to my eye.

I have no doubt that the writer is better at translating than AI, but I have to say that AI translation has gotten so good that I'm not sure how much longer translation work will be there, or rather it might end up being more about auditing.

For example, I just read the Lawrence Ellsworth translation of The Three Musketeers, which I very thoroughly enjoyed. I don't speak or read French, but from my understanding Ellsworth's translation is considered one of the more accurate translations of the work.

Out of curiosity, I sic'd Claude Fable on the original French version of The Three Musketeers and told it to translate accurately, but also try and keep the same jovial tone as the original and do not censor anything. After it was done, I didn't read the entire output, but I did compare a few individual chapters between the Ellsworth translation and the Fable translation.

They were honestly remarkably similar. As far as I could tell, nothing was substantially different from the Ellsworth translation and the Fable translation. I do think that the prose for the Ellsworth translation was a bit better, but the prose for the Fable one was actually perfectly readable. Again, I don't speak French so I cannot say for sure, but I do not believe that I would have gotten a significantly different experience had I read the Fable version instead of the Ellsworth version.

Now, it's possible (and likely) that this is somewhat self-fulfilling; Fable might have been trained using Ellsworth's translation and as such it's very directly able to crib from it; sadly since I do not speak any language outside of English, there's sort of a catch-22: the only way I can compare the accuracy of a translation is to compare against other translations, but if other translations exist then that will likely influence the results, and if a translation doesn't already exist then I have no way of auditing it.

I'm still going to continue reading through Ellsworth's translations for the subsequent stories simply because that feels more canonical, and as I said I do think the prose was a bit better.

You'd be laughed at if you said that ChatGPT could help you with graduate level mathematics in 2024, but this year, AI models on simple prompts are solving previously unsolved Erdos problems.

It seems silly to imagine that there is some fundamental barrier between human intelligence and AI, and that AI could never do many of the things that humans can do. Inferring intent, gauging sentiments, factoring in cultural values, etc. all the things cited as stuff humans can do but AI can't, AI can currently do if given enough context. But more importantly, all those things aren't magical tasks that can only occur inside a human skull, they are a product of information processing, its just the information processing that has been hard to make computers good at, but so far it appears AI keeps getting better.

I'm all for humans having special value that is not attached to their ability to perform useful work. However denying the abilities of AI models seems to be a common mistake many people are making, and sadly reality catches up to these people before they can emotionally prepare.

I think it's an interesting perspective, because translation is one of the jobs that I (a) hear is the first to lose work due to AI, and (b) often used as an example of "acceptable" AI by people who are skeptics of LLMs and AI-generated art.

Everybody else is a fungible cog.

TFA is a good little read - couple things come to mind

  1) Knoll’s Law [0][1]

  2) The ways I feel when I’m working on a hard problem in an area of my expertise and  some person starts in with “Why don’t you just…”. Enough people have come to me in such a situation with such a comment that I think it mostly translates as a sort of shibboleth for “I have no real idea what I’m talking about.” Now to find out if this is a teachable moment, if I have to maintain a sense of humour, or find out if I’m actually one of the days lucky 10,000.[2]

[0] https://effectiviology.com/knolls-law/

[1] https://en.wikipedia.org/wiki/Erwin_Knoll

[2] https://xkcd.com/1053/

Out of curiosity, I pasted an article in French I was reading a few minutes before coming across this thread into ChatGPT and asked for a translation into English. It was certainly passable from a functional perspective, and I wouldn't hesitate to use it to translate an article from a language I don't understand. But it was not professional-quality work. There were a couple instances where the French grammar was mistranslated, and the writing was perfunctory, not going into any effort to have the article flow like it was originally written in English instead of simply translating each sentence literally. Would I read an article written like this? A short one. A novel? Definitely not.

I don't see LLMs being able to replace translators for less-spoken languages.

I know a translator between two Eastern European languages, and some jobs require use of specialized dictionaries. Using LLMs in such cases would be very unreliable and would require even more effort to check and correct than doing it correctly in the first place. Plus, I really doubt that US tech firms are training LLMs on language spoken by "only" 6 million people.

As for entertainment, anyone who grew up in Eastern Europe with pirated movies with nasal monotone translations, or machine-translated video games knows how much those take away from the experience. Sure, "AI could do better", but could it be consistent and capture cultural nuances and idioms, etc?

"we all more or less look the same in gym clothes"

Maybe my brain works differently than the author, but I'm surprised at this statement. Gym clothes don't change recognition for me, it's about the face, body, posture, clothes don't really enter into it. For me it is nonsensical enough to be suspicious.

And for a human centric perspective, not recognizing who someone is sad, it's knowing that you probably won't meet them again so it's not worth it, the community isn't there. Where community and interpersonal relationships between people are something we still hold dearly.

The most important thing a human translator does is certify that the translation is faithful.

Period.

You could do a machine translation if you want, but you better pore over every word in case you end up on the witness stand.

I recently saw a video showing the french to german translation of a french McDonald's terminal. The translations were hilarious bad, like old school google translate bad.

Maybe McDonalds is big enough to not care about their reputation, maybe they are happy about the free clout from people making fun of them but they certainly chose to cheap out on translations.

https://www.tiktok.com/@denneshow/video/7522160205501566230

What’s unfortunate is that the market that is willing to pay for high-quality human translation has shrunken considerably.

I say it’s a simple value proposition.

A few examples

Audio book narration. Human narrators are paid a seemingly ridiculous amount of money to literally read a book out loud. We have the tech to replace them, it’s actually pretty dang good, and it is substantially cheaper to do with computers. It’s pretty accurate too. In the audio book industry though, if you take your book seriously you have a real person read it. The best one you can find that you like. Readers enjoy hearing good narrators and the total value one narrator can bring is very high mostly because the value scales well.

Another real world example that doesn’t scale well, call centers. Customers want humans, but execs have tried to replace them with automation in every way possible. The margins of a business get squeezed because the value of the human touch doesn’t scale well in this case.

Translation falls a bit in the middle. I’m sure ChatGPT is good enough for some people. If you are a restaurant and need to understand what you are ordering at the local authentic Italian restaurant it’ll do the job. If you have a bad food allergy? Maybe not, you are willing to pay for accuracy because that’s what a human brings

So the answer to the question posed in the article, can’t you just upload it to ChatGPT? Maybe yea maybe no

I worked at large Japanese bank in New York and happened to sit near Chief US Economist next to his Japanese translator. She would occasionally ask about certain idioms. I remember explaining what a wildcat strike was for instance. But it must have been pretty tough because the guy was prolific in his commentary.

I had transliterated lyrics of a song * with stanzas in Urdu , Braj Basha, Persian and Arabic , that I wanted to understand better ..

Gemini did a pretty good job of translating this to English .

Sure a professional human translator would have done a more nuanced job if I was willing to invest the money and time . But ...

* tajdar e haram originally by Payam Saihalwi, later versions by the Sabri Brothers and recently by Asif Aslam

So i assume this post is just a bit of writing out frustration, but i'm always hoping that "AI can't do it" posts to include examples.

A list of "Examples AI will silently fail at" would be a lot more interesting, and might just convince your next potential client to _not_ use AI.

As a public service employee within the GOC, I feel the pain expressed by the author. I sat through a meeting today where somebody with no domain knowledge puffed up their chest to show off their gpt created master lesson plan for a four year long internal training plan that is being re-worked.

I could feel the heads of those around the table that had been teaching this material for a decade starting to explode as this was exactly what others in the thread have described: it looked good until vetted by experts, then it was easy to poke holes as it was just not right

The problem in the public service is that the experts who can review the output are leaving or being nudged out.

Sounds a aweful lot like the kind of things we were all saying before realising that we had to change what our jobs meant.

> If you ask me, nothing can save downtown Ottawa or North American public transit.

Come to Montreal. Only 2H away and you can get by decently well without a car.

I'm gonna sound a bit like the clueless gym hr lady: I assume most income generating translation jobs are either mandated by law or commercially high stakes enough to warrant a human to do, no? Were people really being paid to do the type off _low stakes_ translations implied that a automated system can replace?

Maybe a publisher will replace the translator of the next Dan Brown best seller with Mythos? Who cares other than those buying it, getting money out of it?

One of my parents tried this to beat a deadline for product packaging.

There are now bags being sold marked "Lawn Suits", when it was supposed to be Lawn Topdressing

Presumably the people paying the author for translation services are aware of AI, but for whatever reason are choosing a humans services instead. IMO it would be a form fraud to heavily rely on AI and not disclose that to the customer.

unfortunately this person will soon be unemployed.

not because their skills are no longer relevant, but because they are taking a principled stance defending now irrelevant skills.

I can’t believe this article hasn’t been written by ChatGPT. The author claims to have written it but has clearly become completely captured by the awful generic style of AI writing.

> Should you pay your roofer less because he uses a hammer instead of his bare hands?

Yes. Effective tools increase the supply of roofs made. More supply means lower prices per roof. But because the same number of roofs need to get worked on, the increase in roofs per roofer means less roofers will be needed.

    AI isn’t replacing me. Like a toddler, it
    needs to be constantly coached.

Like a toddler, it will grow up.

Humans are really bad at noticing trajectories. They see the current situation. They know what the situation was 5 years ago. But for some reason they do not believe that there is a trajectory. They view the present state as the final destination.

It's quite ironic as the transformer architecture that powers most generative AI was invented for language translation :)

LLM's are in fact very good at translation and transliteration.

Safe to say OP just does NOT like AI https://correresmidestino.com/sorry-i-was-busy-unfucking-my-...

Poor woman should really look into pivoting her career or finding a different way of making money. Truth be told, her industry/career is not going to get better. Consistent work will just not fall from the sky.

Being bitter will not improve her situation. Even organizations like UN/OECD are looking into implementing AI in various ways.

Really good blog though. I love life blogs like these! You can go back and live through so many interesting/pivotal moments.

Denial is tangible

> “Great. So, do you use AI a lot at work?”

> “Oh, I can’t! It’s really not reliable enough.”

Gell-Mann Amnesia strikes again.

As a former freelance translator (1986 to 2005, Japanese to English), I have much sympathy for the writer. But I wouldn’t be so confident that AI cannot do professional-level translation.

She writes: “I adapt, I localize, and I find the best way to convey the original message so it makes sense and feels natural. I research terminology. I make sure it’s consistent throughout.”

I’m sure she has other important insights into what enables her to do her job well. The problem is whether or not such insights can be incorporated into an AI-driven translation system, too.

Since early this year, I have been experimenting with a variety of agentic systems for language-related tasks, including dictionary-writing, research on topics in the philosophy of language, essay-writing, and translation. Other than the dictionary [1], I am keeping the results private, so they haven’t been evaluated by others. But my personal assessment is that agentic systems given suitable high-level guidance can be very good at such tasks now.

If I were still freelancing and I had a large translation job to do for a client, here is the outline of the prompt I would give to Claude to get it started:

“Use this private GitHub repository to build a system for translating [genre of text] from [Language1] to [Language2]. The directory samples/ contains examples of the type of document to be translated, high-quality human translations of those documents, and texts in [Language2] that are in writing styles that I believe to be appropriate for this genre of translation. The file guidelines.md contains my general instructions about the needs of my client and my preferences for how you should translate texts along various axes (natural vs. literal, informal vs. formal, preferred dialect in [Language2], consistency vs. variety in terminology translation, etc.). Begin building (1) a knowledge wiki for this project using Karpathy’s LLM-wiki framework and (2) a system inspired by Karpathy’s Autoresearch, AutoResearchClaw, etc. for testing and recursively improving both the functioning of the system and the quality of the translations. For the actual translation, editing, checking, etc., use not only your own ability and the knowledge assembled in (1) but also outsource such tasks to other frontier models through OpenRouter, and use adversarial evaluations among those models and yourself to check and recursively improve the system design, the prompt-writing for other models, and any translations created by the system. My OpenRouter API key is available in this environment. You may spend up to $xx per day in API calls until this project is ready to do real translations; before beginning a real job, give me an estimate for how much the API calls will cost for that job. The initial build-out of this project will take many sessions, so write a prompt called resume-prompt.md that I can point you to at the start of a scheduled Routine to have you work on this. Commit and squash-merge to main at the end of each session. I will be checking in occasionally to view your progress and to ask you to run translation tests, and I will offer guidance then on how to improve the pipeline further and make the translations closer to what my client needs. If you have any questions before you begin, please ask me.”

[1] https://www.tkgje.jp

Reminded me of this quote:

"Expertise in one field does not carry over into other fields. But experts often think so. The narrower their field of knowledge the more likely they are to think so." - Robert Heinlein

In this case, the gym buddy doesn't think that she's an expert in the other field, but dismisses it as something ChatGPT can do with ease.

the version of this skillset that stays employed is "now I translate 10000x more than i could before by managing a fleet of agents. by encoding my experienced taste and judgement into robust evals, I've helped my ai translators be far better than chatgpt on its own, and much more cost effective compared to manual human translation"

Denial isn't just a river in Egypt

—

Who is gonna tell her?

unfortunately this person will soon be unemployed.

AI should be used for all the bullshit tasks that no one wants to do. There are garbage dumps full of stuff that can be reused and recycled. But it's not high enough ROI to pay someone $25/h to sort trash, so it isn't happening.

You don't even need to argue that you're better than the AI. The point is that the client could have uploaded it to ChatGPT too. Perhaps they even did, and they didn't like the answer they got. They are sending it to a human because they want a human to do the work. If you were to send back ChatGPT output, that would be fraud.

wrt. the end of the story, it will be interesting to see if people start noticing their Dunning-Kruger bias as a result of LLMs.

Specifically: LLMs make it really easy to misunderestimate the complexity of fields other than your own. (You can see this with a lot of vibecoded projects, for example – once they hit the wall of complexity, they stall out or start finding ugly patches for fundamental design issues, etc.)

I don't think this sort of cultural change will happen short-term, though.

Translating is one thing that artificial intelligence undeniably excels at, and the value of this alone is enough to underpin the trillion dollar valuations of the gigantic AI companies.

Translation is a gigantic boon for business, but just as important for human connection, for culture, science, art, and entertainment. The value of automatic and cheap translation between all languages, this tower of Babylon, is immeasurable.

Human translators will always be better than any AI at their job. But they don't have unlimited time and energy, and they aren't cheap. AI makes good to great translations available to everybody.

Any expert in any field will gladly tell you that ML sucks for specifics of their field (and it does). But if you are not an expect in that field, it looks convincing enough to make you think that maybe it is OK for that field, and your field is somehow unique. It is not. Any expect in any field will confirm to you that ML produces plausible-looking slop which is occasionally completely wrong. This is the case for all fields.

This is all bullshit. I speak 4 languages, 3 fluently. Even chatGPT does a stellar job with translation. For most things people want translated- forms, administrative documents etc. I doubt you even need a human in the loop.

That being said, something with essence like a novel definitely still needs to be done by a human.

Jesus fuck, stop with the chatgpt written posts.

[dead]

[dead]

True, and relevant (I live with a professional editor)... yet I immediately think of Ximm's Law:

Every critique of AI assumes to some degree that contemporary implementations will not, or cannot, be improved upon.

Lemma: any statement about AI which uses the word "never" to preclude some feature from future realization is false.

Lemma: contemporary implementations have already improved; they're just unevenly distributed.

This is just about the worst career you could be in right now. Of course people are just going to upload it to ChatGPT. Processing text is its forte.

This person is in the first stage of grief (denial); artists are several stages ahead. Most customers are not going to care about the difference in translation quality unless it's in a regulated sector.

From the post:

> Ah, you can’t fire me, I’m self-employed!

I don't understand thinking like this. I think companies can certainly fire their contractors.

All I got out of this article is that he should have went home and dumped it into chatgpt just to see what happened; then if it did as good a job as him, he should start looking for other places he can add value that AI can't.

I agree with the take, but it's a temporary one, the sad reality is that we will be literally inferior soon, there will be a point where we will not trust human input without counter check by AI, we need to remember that we are kinda at the beginning of the AI era, in 5 to 10 years it's very unlikely that a human translator or software engineers will do better than the tooling we will have.

There is already a tipping point now in software engineering where we prefer to ask AI instead of humans because we believe accuracy will be better, see SO death as an example or just see the current state of online dev communities, it's getting deserted and between team members at work, we can also notice that people speak less and less.

Sad but I believe it.