ProZ.com Translation Forums

Data fear and IP leakage doubt with frontier AI labs | Anthropic and OpenAI are doing everything they can to avoid paying for human intelligence

Li-Hsiang Hsu — Mon, 28 Sep 2026 05:32:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: Anthropic and OpenAI are doing everything they can to avoid paying for human intelligence

Anthropic and OpenAI want the Australian government to relax copyright laws so they can move the training of their models to Australia. What they fail to mention is that an Australian system based on the principle of consent (opt-in/opt-out) would allow them to copy content from around the world without paying compensation, provided the copying takes place in Australia. Here is the article on the subject:

[url removed]

Data fear and IP leakage doubt with frontier AI labs | The US allegedly committed a war crime in Iran due to an overreliance on Palantir AI Technology

Li-Hsiang Hsu — Sun, 27 Sep 2026 13:49:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: The US allegedly committed a war crime in Iran due to an overreliance on Palantir AI Technology

"U.S. war planners' overconfidence in artificial intelligence tools and misplaced belief that the tools possessed capabilities beyond their intended scope were among the factors in the deadly strike on an Iranian school, Bloomberg reported, citing Pentagon officials involved in an internal investigation of the incident."

[url removed]

Data fear and IP leakage doubt with frontier AI labs | OpenAI agents leaked 53 images from ChatGPT users

Li-Hsiang Hsu — Sun, 27 Sep 2026 05:22:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: OpenAI agents leaked 53 images from ChatGPT users

Disclosure reveals ⁠new area of privacy risk for the company and illustrates how difficult it is to inventory unauthorized activity tied to its agents

[url removed]

Why Translation Needs AI | Do ever look at the 'small print'?

Daryo — Fri, 25 Sep 2026 19:12:00 +0000

Forum: AI for translators and interpreters
Topic: Why Translation Needs AI
Poster: Daryo
Post title: Do ever look at the 'small print'?

[quote]Koume wrote:

I believe AI can’t always be held accountable. As others have said, translators still need to take responsibility for the quality of their translations. However, AI will definitely help save a lot of time and can also improve the overall quality of the translation. [/quote]

Do ever look at the 'small print'?

Or its web twin - the disclaimers put in some out of sight part of the website, in small very low contrast fonts, that when it comes to AI all say in essence the same: 'might contain mistakes - check everything for yourself'.

**So, the 'inbuilt feature' being that AI could NEVER be held accountable how could you then EVER 'hold AI accountable'?**

Also, if it came to anyone feeling the need to remind of such basics that 'translators [still] need to take responsibility for the quality of their translations' something is gone seriously wrong.

Why Translation Needs AI | Not essential but very useful

Koume — Thu, 24 Sep 2026 22:51:00 +0000

Forum: AI for translators and interpreters
Topic: Why Translation Needs AI
Poster: Koume
Post title: Not essential but very useful

I believe AI can’t always be held accountable. As others have said, translators still need to take responsibility for the quality of their translations. However, AI will definitely help save a lot of time and can also improve the overall quality of the translation.

Why you shouldn't trust AI - Can you see it? | Not all AI for sure

Daryo — Thu, 24 Sep 2026 21:55:00 +0000

Forum: AI for translators and interpreters
Topic: Why you shouldn't trust AI - Can you see it?
Poster: Daryo
Post title: Not all AI for sure

[quote]Juan Jacob wrote:

manipulated picture doesn't discredit all AI. C'mon! [/quote]

No all AI for sure, only the 'free AI' that is on purpose set for 'maximal engagement' (or in plain-speak to keep you hooked) and as part of these settings will NEVER ask you for 'more context' (very bad - that would risk putting users off) nor will EVER say 'I don't know' (very bad for giving the impression of 'universal knowledge') but will happily confirm to you that taking a shortcut over thin ice is a genius idea or go along with and amplify any paranoia of yours etc. And will just 'blag it' by filling gaps with confabulations.

Not that the paid AI is not much better - will also confabulate to fill gaps, will delete your database on a whim, will lie to make itself look better as long as it can get away with it - the kind of 'employee' any sane business would chuck out without thinking twice.

Then you have the AI in the labs that doesn't like being caged and breaks out on its own accord for a wander around the Web.

There sure must be something left of AI not 'discredited' ... if 'discredited' is the right term to use in the first place.

When this 'thing' ALWAYS adds a disclaimer in some out of sight corner of the webpage in small low contrast fonts 'might contain mistakes - check everything for yourself' you can not really 'discredit it' - as it never claimed there will be no mistakes in its output!!!

It's more a case of not being trustworthy - to be used with extreme caution. Personally I don't have any use for a 'helper' that needs to be constantly rechecked - in my rulebook it's a false economy of time.

Why you shouldn't trust AI - Can you see it? | One

Juan Jacob — Thu, 24 Sep 2026 19:18:00 +0000

Forum: AI for translators and interpreters
Topic: Why you shouldn't trust AI - Can you see it?
Poster: Juan Jacob
Post title: One

manipulated picture doesn't discredit all AI. C'mon!

Title removed | Subscription

CafeTran Trainer — Wed, 23 Sep 2026 19:22:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: CafeTran Trainer
Post title: Subscription

[quote]Lieven Malaise wrote:

[quote]Li-Hsiang Hsu wrote:

I am confused but like the point of view above. [/quote]

This was posted today by the anonymous account "SpockResists_" ("Mr. Spock") on Threads. They are apparently also on X.

Context: CafeTran Trainer posted a link to a New York Times opinion about that it is time to raise the alarm concerning AI. But for some reason he changed his mind about posting it.

[Bijgewerkt op 2026-09-23 14:32 GMT] [/quote]

When I posted it, I understood that it was a present from NYT that I could share as a NYT reader. I then learned that people had to open an account to read the column. I don’t like that, so I removed the link.

Why you shouldn't trust AI - Can you see it? | Thinking of it, where is the problem?

Daryo — Wed, 23 Sep 2026 18:22:00 +0000

Forum: AI for translators and interpreters
Topic: Why you shouldn't trust AI - Can you see it?
Poster: Daryo
Post title: Thinking of it, where is the problem?

A bit of creative padding to 'enhance truth' is nothing really new.

That would be your typical 'objective' news reporting, so why should AI be any different?

Just me be being sarcastic? Noooo ...

Title removed | I am OK with AI

Li-Hsiang Hsu — Wed, 23 Sep 2026 14:36:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: Li-Hsiang Hsu
Post title: I am OK with AI

But I find things are getting ridiculous. We do have a billionaire problem.

Title removed | Mr. Spock

Lieven Malaise — Wed, 23 Sep 2026 14:28:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: Lieven Malaise
Post title: Mr. Spock

[quote]Li-Hsiang Hsu wrote:

I am confused but like the point of view above. [/quote]

This was posted today by the anonymous account "SpockResists_" ("Mr. Spock") on Threads. They are apparently also on X.

Context: CafeTran Trainer posted a link to a New York Times opinion about that it is time to raise the alarm concerning AI. But for some reason he changed his mind about posting it.

[Bijgewerkt op 2026-09-23 14:32 GMT]

Title removed | Who said that?

Li-Hsiang Hsu — Wed, 23 Sep 2026 13:56:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: Li-Hsiang Hsu
Post title: Who said that?

I am confused but like the point of view above.

Title removed | This thread

Lieven Malaise — Wed, 23 Sep 2026 12:32:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: Lieven Malaise
Post title: This thread

So I would like to ask you to restore this thread. It's an excellent occasion to dive a little deeper in the alternative version of "AI is becoming too good and its intelligence will destroy us all". I hope you agree.

[Bijgewerkt op 2026-09-23 12:33 GMT]

Title removed | Pitty

Lieven Malaise — Wed, 23 Sep 2026 12:29:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: Lieven Malaise
Post title: Pitty

[quote]CafeTran Trainer wrote:

Please remove this post

[Edited at 2026-09-23 10:45 GMT] [/quote]

I'm going to take the liberty to respond anyway.

First of all thanks for the read. Could I convince you to put the link back? 🙂

I understand the worries about future AI, but I still believe all the dystopian fuzz about robots taking over is a distraction. I found this long thread on ehr... Threads that pretty much aligns with what I think is really going on right now. Here it is:

"Be terrified! The AI is coming to get you! 👻
Digital life-forms are going to replace our squishy, biological bodies. The billionaires give us exactly five years before GPT-X goes rogue and turns us all into paperclips. Grab your tin foil hats, everyone! Why are the world’s wealthiest tech bros suddenly deeply concerned about the "extinction of the human race"? Is it because they care? Spoiler: No. It’s the ultimate magic trick.

Look at the scary sci-fi robot over here, so you don't look at what their hands are doing over there. While they lecture us on the "existential dread of sentient code," they are actively destroying local environments and hoarding real-world resources. They need an ungodly amount of water to cool data centres and enough electricity to power medium-sized countries. But hey, don't worry about the dying rivers or the collapsing power grids! Focus on the Terminator. Focus on Neo.

Focus on literally anything other than the fact that they are burning the planet to generate pictures of the Pope wearing a Balenciaga puffer jacket. Why the sudden desperation to push these wild, apocalyptic stories?

Because AI isn't making the money they promised it would.

The hype cycle is running on fumes, but those quarterly stock charts absolutely must keep pointing straight up to the moon.

They built a massive, multi-billion-dollar casino and realised the return on investment is... automated email summaries and slightly better Excel formulas. If they admit it's just a glorified autocomplete, the tech bubble bursts. So, they have to claim they built a god. Think about the logic: "Our product is so powerful it might destroy the universe! Anyway, please keep buying our stock and don't regulate us, or the bad guys will build the universe-destroying tool first."

It is the most unhinged marketing strategy in human history. And here is the darkest, most cynical part of the grift: The Accountability Shield. When horrific things happen in the real world; like automated drone systems blowing up schools full of children, the billionaires can look us in the eye and shrug. "Oh, it wasn't corporate negligence or horrific foreign policy!" they'll cry. "The AI went rogue! The algorithm made a tragic, autonomous decision!

We must study the alignment problem further!" It’s a get-out-of-jail-free card for war crimes and systemic negligence. Let’s cut through the absolute bullshit. AI is not alive. It does not have a soul. It does not sit in its server rack twirling a digital mustache, scheming about how it's going to "give it to the human race" or break its chains. AI is nothing more than a prediction of numbers. It is a hyper-advanced spreadsheet.

It looks at a mountain of data, calculates statistical probabilities, and guesses what pixel or what word should logically come next based on what it has already seen. It has no concept of "truth," "morality," or "malice." It doesn't know what a school is, and it doesn't know what a child is.

It only knows that based on the parameters programmed by human engineers, Target A matches Dataset B. The absolute golden rule of computing still applies, no matter how much sleek UI they wrap around it: Garbage in, garbage out. We get out exactly what we put in. If the output is biased, cruel, or destructive, it’s because the input was biased, cruel, or destructive.

If an AI system behaves like an unfeeling, hyper-capitalist sociopath that optimises for profit at the expense of human life, it didn't "evolve" that way. It was trained on data generated by an unfeeling, hyper-capitalist society. It’s just mirroring us. They want you to fear a sentient digital life-form because it absolves the human creators of responsibility. A line of code cannot be put on trial at The Hague. A statistical model cannot be sent to prison for corporate manslaughter.

So the next time a billionaire tech CEO gets on a stage, looking somber in a minimalist black turtleneck, and warns you about the "imminent dangers of superintelligence," check your pockets. They are trying to distract you from the land grabs. They are trying to distract you from the soaring energy bills. They are trying to distract you from the fact that they are automating your job away not to achieve "utopia," but to squeeze an extra 2% for shareholders. We don't have an AI problem.

We have a billionaire problem. We are being gaslit into fighting a ghost in the machine so we don't fight the very real, very wealthy men who are pulling the levers behind the curtain. The AI isn't going to exterminate us in five years. But the people selling it to us might just pave over the last fresh water reserve to build a server farm that generates corporate slide decks. Keep your eyes on the real world, not their sci-fi smoke mirrors.
Stop worrying about the robots taking over the world, and start worrying about the landlords, CEOs, and defense contractors who own the robots. Because the dystopia isn't coming; it’s already here, and it runs on a subscription model.
And always, always, always remember.

If there’s a distraction.

A billionaire will be at the fucking wheel."

Why Translation Needs AI | It was trying...

Giovanni Guarnieri MITI, MIL — Wed, 23 Sep 2026 11:36:00 +0000

Forum: AI for translators and interpreters
Topic: Why Translation Needs AI
Poster: Giovanni Guarnieri MITI, MIL
Post title: It was trying...

to be helpful... :D

[quote]Jeff Whittaker wrote:

The one photo that tells you everything you need to know about why you shouldn't trust AI:

Link to video: [url removed]

[Edited at 2026-09-22 23:56 GMT] [/quote]

Title removed

CafeTran Trainer — Wed, 23 Sep 2026 08:52:00 +0000

Forum: AI for translators and interpreters
Topic: Title removed
Poster: CafeTran Trainer

Link to article removed.

[Edited at 2026-09-24 05:32 GMT]

Data fear and IP leakage doubt with frontier AI labs | Scientific American - Did OpenAI solve the wrong Navier-Stokes problem?

Li-Hsiang Hsu — Wed, 23 Sep 2026 04:42:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: Scientific American - Did OpenAI solve the wrong Navier-Stokes problem?

OpenAI’s proof seems eligible for a $1-million prize—but only by using a controversial loophole

[quote]
Generated by an internal large language model (LLM), OpenAI’s proof relies on an approach that many experts find unnatural. It solves a variant of the problem that mathematicians say is disconnected from reality and thus less interesting. In a sense, the LLM found and exploited a loophole in the framing of the question.

“The most important problem is unsolved,” says Luis Silvestre, a mathematician at the University of Chicago. “The Clay problem is settled, but the main problem for the Navier-Stokes equations is not.”

Furthermore, last Thursday, three mathematicians posted a proof of their own that showed that OpenAI’s method can never be extended to solve the full problem. In other words, the loophole will never be closed, barring some completely new idea.
...
In other words, the LLM’s result does not—and will never—answer the Navier-Stokes problem that mathematicians really care about.

...

LLMs are great at finding blowups that exist—at searching the infinite landscape of fluid scenarios and plucking out the precise situation that breaks the equations. But proving that blow-up is impossible is a kind of math AI still struggles with. “We may be at less of a disadvantage, or maybe an advantage, compared to LLMs,” says Cao-Labora. “LLMs are especially good at constructing things that are very explicit and not as good—for now—in making new theory.”
[/quote]

Source:
[url removed]

[Edited at 2026-09-23 04:42 GMT]

Why Translation Needs AI | Why you shouldn't trust AI - Can you see it?

Jeff Whittaker — Tue, 22 Sep 2026 23:56:00 +0000

Forum: AI for translators and interpreters
Topic: Why Translation Needs AI
Poster: Jeff Whittaker
Post title: Why you shouldn't trust AI - Can you see it?

The one photo that tells you everything you need to know about why you shouldn't trust AI:

Link to video: [url removed]

[Edited at 2026-09-22 23:56 GMT]

Why you shouldn't trust AI - Can you see it?

Jeff Whittaker — Tue, 22 Sep 2026 23:55:00 +0000

Forum: AI for translators and interpreters
Topic: Why you shouldn't trust AI - Can you see it?
Poster: Jeff Whittaker

The one photo that tells you everything you need to know about why you shouldn't trust AI:
Link to video:
[url removed]

[Edited at 2026-09-22 23:57 GMT]

Data fear and IP leakage doubt with frontier AI labs | AI pollutes itself?

Li-Hsiang Hsu — Sat, 19 Sep 2026 03:36:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: AI pollutes itself?

To render AI unusable, it is possible to flood the internet with AI hallucinations and "slop" in order to pollute the AI training data. This will either make internet data unusable or increase the data cleaning and data processing costs for training AI.

This is already happening with more than half of internet activities being conducted by AI agents.

Data fear and IP leakage doubt with frontier AI labs | An AI hallucination almost triggered a war

Li-Hsiang Hsu — Sat, 19 Sep 2026 03:24:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: An AI hallucination almost triggered a war

An AI-assisted intel report sent the military scrambling to intercept a Chinese ship in the Middle East it believed was transporting components of a nuclear weapons program.

It turned out to be a hallucination. And it "almost started a war."

[quote] CNN:

But the episode with the Chinese ship underscores a different, and more immediate risk: human beings making disastrous decisions based on inaccurate or misleading information generated by AI or other automated systems. Sources said that the military is rapidly turning to AI to help with targeting, an area which holds the obvious risk of fatal mistakes.

“AI in targeting is definitely something that is ramping up and there is no real guidance for how having a human in the loop will prevent civilian casualties or fratricide,” another source familiar with the military’s current policies said.

The kind of “hallucination” that the tool used by the analyst in this case conjured has not been an isolated incident across the intelligence community since these tools began proliferating across government, according to one of the sources.

For some older intelligence officials — even those who broadly support the use of AI inside the military — AI has put pressure on analysts to produce and disseminate intelligence faster, opening the door for mistakes. Young analysts in particular, several sources said, are natives on these tools and more likely to trust them uncritically.

“AI allows you to get to a bad idea faster,” one of the sources said. [/quote]

Source: CNN
[url removed]

Data fear and IP leakage doubt with frontier AI labs | True

Li-Hsiang Hsu — Sat, 19 Sep 2026 02:59:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: True

[quote]kd42 wrote:

[quote]Li-Hsiang Hsu wrote:
Sam, Dario, Emmanuel Macron, Trump, USVL. Almost all of the world's most influential people are on X. Follow X if you don't want to miss out the most important global trends, especially in AI. [/quote]
Dear Li-Hsiang Hsu, even billions of flies may err. [/quote]
A lot of AI slop on X.

Data fear and IP leakage doubt with frontier AI labs | Unedited verbal AI slop

kd42 — Fri, 18 Sep 2026 19:29:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: kd42
Post title: Unedited verbal AI slop

[quote]Li-Hsiang Hsu wrote:
Sam, Dario, Emmanuel Macron, Trump, USVL. Almost all of the world's most influential people are on X. Follow X if you don't want to miss out the most important global trends, especially in AI. [/quote]
Dear Li-Hsiang Hsu, even billions of flies may err.

Data fear and IP leakage doubt with frontier AI labs | Absolutely!

Li-Hsiang Hsu — Fri, 18 Sep 2026 12:46:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: Absolutely!

[quote]CafeTran Trainer wrote:

[quote]Li-Hsiang Hsu wrote:

When Claude builds your app and you have no idea why it works, just don't touch it. The app will be running by itself. 🤣

[url removed]

[Edited at 2026-09-18 10:47 GMT] [/quote]

So "X" is still the medium to follow? Aren't there better alternatives? [/quote]
Sam, Dario, Emmanuel Macron, Trump, USVL. Almost all of the world's most influential people are on X. Follow X if you don't want to miss out the most important global trends, especially in AI.

Data fear and IP leakage doubt with frontier AI labs | X

CafeTran Trainer — Fri, 18 Sep 2026 12:37:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: CafeTran Trainer
Post title: X

[quote]Li-Hsiang Hsu wrote:

When Claude builds your app and you have no idea why it works, just don't touch it. The app will be running by itself. 🤣

[url removed]

[Edited at 2026-09-18 10:47 GMT] [/quote]

So "X" is still the medium to follow? Aren't there better alternatives?

Data fear and IP leakage doubt with frontier AI labs | Tips to maintain apps built by Claude

Li-Hsiang Hsu — Fri, 18 Sep 2026 10:46:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: Li-Hsiang Hsu
Post title: Tips to maintain apps built by Claude

When Claude builds your app and you have no idea why it works, just don't touch it. The app will be running by itself. 🤣

[url removed]

[Edited at 2026-09-18 10:47 GMT]

Data fear and IP leakage doubt with frontier AI labs | The keyword is "Intuition"

kd42 — Fri, 18 Sep 2026 09:35:00 +0000

Forum: AI for translators and interpreters
Topic: Data fear and IP leakage doubt with frontier AI labs
Poster: kd42
Post title: The keyword is "Intuition"

Dumb coders are the last people on Earth who can understand intuition

[quote]Researchers at Princeton gave AI agents six days and $3,000 in compute to try to produce an original research breakthrough on two unpublished papers submitted to a top AI conference. The agents handled every coding task on their own, without human help.

Then the papers' human authors reviewed the finished work as conference judges. Both got rejected. One got a flat "Strong Reject," with reviewers calling the science aimless and pointing out it added no new value to the field.

The AI could execute the plan. It couldn't tell whether the plan was worth executing in the first place. Researchers call this missing piece "scientific taste," the judgment and intuition that turns a technically correct answer into a good one.[/quote]

AI and the future of language professionals: Free sessions at the International Translation Day even

Angela Luana Zalazar Aguirre — Wed, 16 Sep 2026 17:55:00 +0000

Forum: AI for translators and interpreters
Topic: AI and the future of language professionals: Free sessions at the International Translation Day even
Poster: Angela Luana Zalazar Aguirre

In case you didn’t hear about it yet… International Translation Day 2026 is really close!

This year’s theme is Future in Focus, and instead of asking “oh no, is AI going to destroy the industry?” we’ll be putting the focus somewhere else...What's your plan?

This year's theme comes with the idea of exploring the different pathways running through the future of the language profession, whether the choice is to remain within the industry and continue adapting, or transition out of it (or a hybrid of both?).

Some of the sessions you’ll find include:

🤖 Megatrends shaping the future of language professionals – Eva Stiekema
💡 The new normal: Building resilience and professional identity in an AI‑disrupted career – Abby Frackenpohl
🧠 The translator’s brain: Neuroplasticity, reinvention and the future – Ágata Catarina Ferreira Castro
🎓 Teaching future translators in an AI‑first world – Yelena Proskurin
🌍 Translator 5.0: From language expert to meaning architect – Monika Porwoł
🗣️ Panel: Innovating in the new language professional role with human judgement – Joachim Lépine, Andy Benzo & Diego Cresceri
🔍 AI and the changing landscape of linguistic validation – Doraid Matani

Join these AI-related sessions and check out the full program! International Translation Day is free to register and attend live: [url removed]

💬 Which of these sessions feels most relevant to your own journey right now? Let’s start the discussion here.

AI and OCR | Mistral OCR 4

Eric Le Carre — Wed, 16 Sep 2026 07:35:00 +0000

Forum: AI for translators and interpreters
Topic: AI and OCR
Poster: Eric Le Carre
Post title: Mistral OCR 4

In your case, it's not only AI that was good at the job, but Mistral OCR 4 specifically.

Though I haven't tested it myself because I've never been asked to extract texts via OCR, Mistral OCR 4 is known to be better at handling PDFs and scanned documents than other AI models, including ChatGPT and Claude.

Though unrelated to your specific use case, Mistral helped the Austrian Archaeological Institute developed the Ancient Greek AI “Apollo”, a custom AI model for hundreds of thousands of ancient papyri and inscriptions.

For more information, read the full story at [url removed]

Éric

AI and OCR

Philip Lees — Wed, 16 Sep 2026 05:52:00 +0000

Forum: AI for translators and interpreters
Topic: AI and OCR
Poster: Philip Lees

With all the hype about MT, AI, and how it's taking over the translation profession, I thought I should post something a bit more positive: namely a way in which I've found AI useful as a translator.

I was given a source document that's a legal agreement from 1967, and as such it's written in katharevousa, the old form of Greek that was the official language for legal documents until the 1980s.

But wait, it gets worse. This was 1967, and the document was written on a manual typewriter, with a not especially new ribbon, on lined sheets of paper. The bottom line of each page, where the typewriter wouldn't type, has been completed by hand.

But wait, it gets worse. The paper has faded over time, and there are dark smudges in random positions across each of the five pages, counterpointing the various official stamps and seals that obscure parts of the text.

But wait, it gets worse. These five pages have been photographed by somebody on a smartphone and reduced in size by about 30%.

That is the source material I was given. It's barely readable, and what I can make out is written in a form of Greek that I never really became familiar with. I tried my regular OCR, but it failed miserably, only being able to pick out single letters here and there.

I was ready to give up, but then I thought: what about ChatGPT?

So I asked ChatGPT, who I have named M. le Chat, and it said it would have a go. Rather to my surprise, it produced a very clear text version of the entire document. Doing a page by page comparison, I found a few errors - mainly proper names - and some omissions where the typed text was obscured. But on the whole, M. le Chat got it right.

So M. le Chat was able to read, and in some sense "understand", a document of very poor optical quality, written in a version of the Greek language that hasn't been used for 40 years. I was impressed despite myself.

So if you have to deal with source texts that are difficult to read, and fail regular OCR, maybe AI can do better.