In my Ottawa life, every Tuesday evening, I take two gym classes back to back—boxing and the pompously named “body sculpt,” which makes me discover muscles I didn’t know I had.
It’s fun. I love it.
But a couple of weeks ago, I ended up cancelling my second class—one of those nights when the first assignment landed in my inbox at 4 p.m., another one arrived while I was on my way to the gym, and a third one popped up right as I was standing in the locker room. All due the following morning, obviously. Welcome to the life of a freelance translator.
Work takes priority over muscles. I headed for the lockers at the end of boxing class.
“Are you leaving? You’re always taking this class!”
I turned around. I was changing into my translator clothes—jeans and a T-shirt—and she was presumably changing into her gym clothes, except first, she was busy taking off her jewelry.
Her look was very polished—the kind of polished that screams office day. Over the past few months, the generous pandemic work-from-home policy had been tightened, scaled back, amended and more or less rescinded in a desperate attempt to have employees single-handedly save downtown Ottawa’s many small businesses and general gloom by their mere on-site hot-desking presence.
If you ask me, nothing can save downtown Ottawa or North American public transit.
“I see you there every week!”
Apparently, I owed her an explanation and possibly an apology. I didn’t remember her, but it’s a very full class and we all more or less look the same in gym clothes.
“I’ve just received some work,” I explained. “I’m a translator and I have three deadlines by tomorrow morning, so I should probably get started.”
“But… it won’t take long. Don’t you just upload the documents to ChatGPT?”
I paused for a split second. Surely, she was joking.
I looked up at her.
She was not.
“It… doesn’t exactly work like that.”
“You should try it, it’s so much quicker!”
Oh. My. Fucking. God.
But hey, I parent a teen. I can recognize a teachable moment when I see one.
“It’s not that easy, you know. Technically, ChatGPT will spit out a translated document. But first, there may be formatting issues. And most importantly, the translation will be questionable.”
“Why?”
“Because AI isn’t human, and it takes an actual person to understand what another human is trying to say—and how to say it so someone else understands it. I don’t just make grammatically correct sentences in another language. 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 sorry, I’m better than AI.”
We’re all better than AI. AI is just better at pretending it can do the job.
Go ahead, ask me how I know.
Yes, obviously, I tried translating with AI.
Ah, you can’t fire me, I’m self-employed!
I’ve been playing with AI since the fall, when it started stealing my job for real. I could either declare it evil and turn into one of those people who will never get a smartphone, or use it to my advantage.
I’m practical. I chose the second option.
AI can’t translate for me. It can’t write either—unfortunately, ChatGPT can’t vouch for the fact that this article is my idea, that it’s my gym, my ignorant civil servant and my punchline. Just take my word for it, pun slightly intended.
And while this article is written by yours truly, you bet I’m going to spell-check it. I probably won’t use AI; I have Antidote. But maybe I will ask Claude’s opinion, and if one of the suggestions is smart—cutting a paragraph, for instance, or clarifying a sentence—I might accept it.
When I started translating 15 years ago, we used to paste uncooperative sentences into Google Translate to see if it had interesting ways to phrase things differently. Then came DeepL—same idea.
What do you think? That we’re translating with pen and pencil? That your accountant doesn’t use fancy Excel formulas? That your manager formatted the PowerPoint alone? That your favourite restaurant doesn’t Google trendy recipes?
We are professionals using tools.
But that’s just what they are—tools.
One of my clients has insane style guides, plural. I’m talking about 500-page documents detailing the proper way to format quotes and the one true way to insert footnotes. I fed them to ChatGPT for the final checks—it can kind of flag when I break a rule. I’ve also used AI to extract specialized terminology from reference documents and build my own glossaries. It’s faster than Ctrl+F, and less likely to make me scream.
But everything has to be double-checked, triple-checked. It’s another way of working, not a magic button.
AI isn’t replacing me. Like a toddler, it needs to be constantly coached. It invents acronyms and organization names, forgets to translate entire sentences, ignores the provided terminology unless repeatedly threatened, and occasionally misses the point completely.
Which is why we—translators, writers, editors, and other professionals—shouldn’t suddenly be paid less because AI exists. Should you pay your roofer less because he uses a hammer instead of his bare hands?
But judging by her amused smile, my civil servant wasn’t getting the point.
“But AI is getting better all the time!”
“What do you do?” I asked, changing tack.
“I’m the Director General, Human Resources and Corporate Services, but I’m currently in an acting position for Workforce Planning and Resources Management.”
This actually made sense to my Ottawa brain. Told you, I’m a translator.
“Great. So, do you use AI a lot at work?”
“Oh, I can’t! It’s really not reliable enough.”
For fuck’s sake.
And she works in human resources!
“AI isn’t replacing me. Like a toddler, it needs to be constantly coached.”
AMEN! AI is a tool, and whilst I would’ve given my right arm for it when I was a Technical Writer for the same reasons you expand on, it would still require me to do the work of learning the application, writing the instructions, testing them, amending them…
The hype is too big, the crash will come, soon?
I hope you’re no longer a technical writer because you choose to move on to another adventure, and that AI didn’t steal your job! I don’t want to read or use technical writing “invented” by AI.
It should crash. But I’m worried about the long-lasting damages to the workforce. So many jobs were devalued because “eh, it’s not rocket science, AI can do it!”
Unpopular opinion= i hate ChatGPT. AI has infiltrated the accounting field. Some are good and some not great and can have detrimental effects to companies. I like AI productivity tools, but i only use a few. I am waiting for the AI bubble to burst patiently…
It’s a popular opinion with my people!
I believe AI *tools* can be useful. It still takes a brain to use them, along with expertise, judgment, etc. I certainly won’t do blind accounting with AI, I’ll keep my human accountant.
AI is great at taking away _your_ job but it’s not reliable enough to do mine is the Knoll’s Law / Gell Mann Amnesia effect of AI.
Lovely write up.
Thank you!
I have to look it up, thank you for the insightful comment.
[…] 原文链接：Hacker News […]
[…] A verdadeira inovação não reside em enviar arquivos para uma interface, mas em construir sistemas que processam esses dados antes mesmo de chegarem ao modelo. A arquitetura de um sistema robusto deve incluir etapas de pré-processamento, como limpeza de OCR, extração de entidades e estruturação em JSON. O artigo original, que discute as nuances dessa abordagem, pode ser consultado no Artigo de Origem. […]
Holy crap I laughed hard at “Oh, I can’t! It’s really not reliable enough.”
This is a true story and when she said it, I was hesitating between laughing and face palming 😆
[…] 来源：Hacker News · 24h最热 […]
Does the country of the AI’s origin make any difference in translation quality? Let’s say I want to translate between English and Mandarin. ChatGPT was built in California, of course. China has AI models too, such as Z.ai GLM-5.1. Would using Z.ai produce a better result because it was built by native Mandarin speakers and perhaps trained on more local resources? I wonder if anyone has ever tested this.
This is an excellent question, and I’ve been wondering the same!
I briefly tested on Chinese and I found ChatGPT was very “international Chinese”, not “mainland Chinese”. I.e. no slang, for instance.
Generally speaking, I find it has a North American biais. For instance, if I ask questions about food (in either French or English), the suggestions are always very North American–peanut butter, Greek yogurt, almonds… not the kind of snacks an average French would reach out for.
I have to try Z.ai.
the chinese version will be better as it is trained in much more chinese language than a NA one.
I speak japanese and z.ai and chatgpt have differences when it comes to street slangs in articulating something.
So it is a matter of training rather than geolocation. Both will produce an understandable version no doubt.
The truth is that translation is dead. I will take a book in any language and AI will devour it into details in seconds. As to a very high level of translation of a sentence, it can be reached, through multiple iterations, and styles that one wants. I can get a Dario Fo style or Bill Maher, applied to a translation or creation and amazing enough the results are incredible.
While keeping an eye on the flow, this can make wonders.
Understandable, yes. Enjoyable? I’m not convinced.
AI translation recognizes patterns, and that’s exactly the problem: after a few paragraphs, it often starts to feel repetitive. Maybe even sooner. It has this slightly mechanical aftertaste, even when it’s technically correct. And of course, it can still be hilariously wrong.
I don’t think translation is dead. I think bad translation is more exposed now, and average translation is easier to produce. But good translation is still a creative act. It’s not just transferring meaning from one language to another; it’s making choices about rhythm, tone, intent, humour, cultural references, ambiguity, and voice.
AI can help, sure. It can produce something understandable. But “understandable” isn’t the same as alive.
Embrace AI, sharpen it as a tool and believe me you will be better off.
Let me put it this way: if someone hired you when you were typing code character by character, imagine what you’re worth now that you can deliver in an hour what used to take a month.
Technical skills were never really the moat. They were the entry ticket. The moat was always the judgment, the relationships, the ability to ask the right question in the first place.
it is a mindset thing.
I completely agree with you.
I went from “this shit is dumb” to “let’s use it as a tool”, and now I’m finding great value. I build glossaries, check references, proof, and more. But I can’t just tap the magic button and produce a translation that works 😉
[…] Source: "Don't You Just Upload It to ChatGPT?" […]