“AI took my job, literally”—Gizmodo fires Spanish staff amid switch to AI translator

Meanwhile, readers say that some AI-penned articles switch languages halfway through.

See full article...

 

[drinkingcoffee]

Google Translate went online 17 years ago -- are the relatively new AIs really better at translating than Google with almost two decades of practice? If not, then I guess I'm curious why such transitions didn't happen years ago.

Unfortunately, I'm not fluent enough in a second language to make any sort of judgement call on AI versus Google Translate

Deepl is, and has been, much better than Google translate for a while (especially at Spanish, in my experience). Which makes it strange/funny that they are using Google translate. None of these tools are good enough for publishing on their own.

Regardless... What's the point? Honestly, automatically translating your website for me is lame. If I need that, I can have my browser or a plugin do it for me, and it will provide more options. If you're not providing additional, original, local content, (or at very least having someone edit the translations), it's worse than doing nothing in my opinion.

Who is their market now? Non-english-reading Spanish speakers, who don't know about automated translation (something that Google Chrome pushes on you unless you disable it)?

 

[windnwar]

I'm fully expecting these companies to then hire gig workers via task rabbit or something similar to go back and fix the mistakes, of which they'll pay them pennies on the dollar for along with no benefits and the proclaim how much money they've saved for stock buybacks at the next shareholder meeting.

That pretty much seems to be the approach to things these days. So what if it just makes the overall population more poor and vulnerable.

 

[LuDux]

Keep in mind you don't own copyright to AI creations. Gizmodo just shit themselves in public with this.

 

[Obsti]

Can we please stop using that word for literally everything? It's supposed to describe a rather specific pattern of degradation that doesn't really explain or fit things like Gizmodo. Gizmodo is an example of something dying just because it is, the content didn't keep up with the competition and started to lose traffic/money etc, and so classic dumb business moves to penny pinch happen. That's very different from the enshitification cycle of trying to transfer the value surplus of locked in users/vendors back to the platform and making everyone else miserable

I take it you are unhappy with the enshittification of comments?

 

[fryhole]

At this point I feel that one of OpenAI's competitive advantages is that they have a snapshot of the Internet from before it was corrupted by LLM hallucination articles and translations.

Perhaps the Internet Archive can make some money by selling Wayback Machine access to AI companies.

Most of the wayback machine is broken. So a bad idea to build anything viable from that data. As groups find this out more and more The IA won't bank too much on that revenue stream - unless the owners are sly enough to dupe some early adopters before they realize anything.

As for Open AI - don't fool yourself; experiements across the web were going on long before they showed up in the public spotlight - so likely some of their data is already hosed. And companies like Google have taken snapshots of their data sets (and retained them) from Day 1 - so don;t think MS or Google can't go back and look at historical data at points in time long before even OpenAI became a thing. Google has a decade on them and MS predates the interwebs.

Your premise is akin to someone saying Apple invented MP3 players or smartphones (hint: they didn't; just as OpenAI is not the only company that has "uncorrupted" data sets)

 

[fruitbane]

My hope is that advertisers will not want their content associated with awkward-at-best and offensively-incorrect-at-worst generated content and that G/O Media will realize their best path forward is either through human-created content or to just shut down and save everyone further pain.

 

[goatsitter]

Google Translate went online 17 years ago -- are the relatively new AIs really better at translating than Google with almost two decades of practice? If not, then I guess I'm curious why such transitions didn't happen years ago.

Unfortunately, I'm not fluent enough in a second language to make any sort of judgement call on AI versus Google Translate

As someone who routinely translates German and French into English and vice versa as part of my job, I can confirm that automatic translation improved a lot. The switch to machine learning led to amazing results. Especially in the last five years it made huge advances. At least for those three languages Deepl is the best right now, but even it does not replace a human translator yet, but it does a lot of the work for me. I am about seven times faster than ten years ago. Now it's more or less as if I was proof-reading an inexperienced translator.

 

[bobbed]

Yes. Significantly. Try using DeepL. I'm shocked there are still many human translator left.

One of the many human translators left here (hello! ).

DeepL is indeed top notch, as is also Google Translate and a couple of other very interesting Neural Network-trained Machine Translation engines (ChatGPT has actually been a bit worse in my experience). At least for the language pairs they’ve been trained directly (English is used as the middle-language when not), they produce great, mildly context-aware translations most of the time—and even some incredible output every now and then.

They’ll still shit the bed often, though, both because no amount of training makes anything perfect and also because it’s sort of a GIGO situation with source text quality (and type; not exactly the bee’s knees at poetry, for instance).

As with almost all AI-driven innovation in almost any sector, unless your need for the final product is merely borne of curiosity or the need to get the gist of what you’re seeing (i.e. not the whole picture), human in the loop is still the way to go—as I hope the brass at Gizmodo finds out soon and with extreme financial and reputational prejudice.

EDIT: grammar + specified the good Machine Translation engines I was referring to are the NN-trained ones.

 

[KingKrayola]

I take it you are unhappy with the enshittification of comments?

Some of us have known that everything is going to shit since before it was cool.

We don't need no stinkin' neologisms.

 

[Honeybog]

Way back in the day, I remember “winning” a few Gizmodo Photoshop contests for sending in some really dumb and bad photoshops of, like, Palm OS and getting a star (basically keeping all of your comments on the top of the page) for some comment about 3G coverage in Manhattan.

I’d like to say that this news makes me sad, but Gizmodo has been dead for at least a decade.* With the exception of Ars, it really seems like tech press has been moribund from about the time Engadget became a PR wire service. El Reg at least still has some fun comments.

*Along with the other Gawker sites, which pretty much became unreadable before the Hulk Hogan lawsuit (with the exception of Deadspin, which hung on to the bitter end). It’s a shame they somehow managed to bring down the AV Club as collateral damage, though.

 

[Honeybog]

Some of us have known that everything is going to shit since before it was cool.

We don't need no stinkin' neologisms.

“The real eternal September happened when ARPANET expanded to the East Coast!”

 

[S-T-R]

with the exception of Deadspin, which hung on to the bitter end

They never quit.

 

[cobaltalpaca]

Google Translate went online 17 years ago -- are the relatively new AIs really better at translating than Google with almost two decades of practice? If not, then I guess I'm curious why such transitions didn't happen years ago.

Yes. 'AI' (LLMs) will put together coherent, grammatically correct sentences (sometimes better than an average native speaker) that aren't a pain to read.

But as with all LLMs at the moment, they have a pretty big chance of those coherent sentences starting to go completely off-topic and bizarre.

I can only speak for Japanese, but LLMs can certainly put out something pretty decent that could just about be used as is. They are better than translation tools which within a sentence or two will have mistakes.

But as I said, you need to check if the LLMs are giving you grammatically correct gibberish, or worse. Translation tools stick to what you input, even if they sound odd or are using incorrect grammar.

 

[Mute999]

I have no evidence to back this, but I feel like English is an awful starting point for any automated translation effort, as it lacks many of the nuances other languages seem to have?

I’m a native Portuguese speaker, and text automatically translated just feels… wrong. Given the choice I would rather read the original.

I bet it’s the same for Spanish.

I have never liked this take on English. I am not a native English speaker either, but every time I heard people make claims of how nuanced and rich their language was compared to English (often Francophones making this claim in my case), I found it was mainly their grasp of the English language that was lacking, giving them the impression English was a more simple language. English is simply the main language used on the www, by many speakers that are non-fluent, leading to a lot of text that is relatively simple. But reading a large number of books in English has shown me that it, too, is very rich. That kind of prose is just not the stuff you'd get exposed to on the internet unless specifically seeking it out.
Agreed on reading the original language over a translation whenever you can, though.

 

[cobaltalpaca]

I have no evidence to back this, but I feel like English is an awful starting point for any automated translation effort, as it lacks many of the nuances other languages seem to have?

This is a common and seemingly deep rooted myth or misconception. Or worse, but let's not go there.

There is completely of gibberish in English that's been spat out by translation tools and now LLMs. Stuff that annoys for exactly the same points butchering of Portuguese annoys you.

 

[lobabobloblaw]

I expect only more of this moving forward.

I wish I didn’t, but I do.

Organize in person, and have discussions about AI as such. Do science pubs, and spread the word about what it can, might, or will do—for good or for bad—but do it in person.

 

[jhesse]

No, because EVERY SINGLE PERSON WHO USES THAT NON-WORD THINKS THEY'RE THE BLOODY CLEVEREST BERK IN THE WORLD. [citation: every single time it's used]

But it's just so... cromulent.

 

[Donkey Kong]

Modern AI translation is pretty good, I could see downsizing your translation staff to just one person who feeds the machine and does a quick QA pass on the results. It's sad to see a rich international presence get the axe, but if the money has dried up and your website is dying and it's between cutting back to all english or using a machine assisted skeleton crew, then so be it, I'd take the latter.

But everyone seems to want to set the machine at a big pile of data, NOT CHECK ITS WORK, and shove the content out the door. I can't even come up with a single word for this! Stupid, short-sighted, irresponsible, self-sabotaging.

I'm honestly more confused than anything because it should FEEL wrong to the website. They should feel it deep in their bones that this process is not adding value to their content. In fact it is subtracting value.

If users throwing a URL into google translate are getting a similar experience with a similar guarantee of accuracy, then you have failed to provide anything new to your readers. Generally when you realize you're doing something pointless, you should stop.

 

[Edgar Allan Esquire]

Some of us have known that everything is going to shit since before it was cool.

We don't need no stinkin' neologisms.

“Our sires’ age was worse than our grandsires’. We, their sons, are more worthless than they; so in our turn we shall give the world a progeny yet more corrupt.”

Horace, First Century Before Cool

 

[Alejandro Nova]

Just read this and I, as a native Spanish speaker, was completely baffled.

English: "Three Decades After Launch, Microsoft's WordPad Is Headed to the Trash Bin"

Gizmodo auto-translation: "Tres décadas después del lanzamiento, WordPad de Microsoft se dirige a la papelera"

My translation (just for you to see the difference): "Tres décadas después de ser lanzado, Microsoft envía su WordPad a la Papelera de Reciclaje"

"Papelera de Reciclaje" is what the Trash Bin is called in the Spanish Windows localization since Windows 95, so you can understand it. The AI version is missing all the historic cues of the original and all of its subtleties. So we can send the AI version and the whole half-ass-translating AI to the Trash Bin, too.

 

[onlysublime]

Google Translate went online 17 years ago -- are the relatively new AIs really better at translating than Google with almost two decades of practice? If not, then I guess I'm curious why such transitions didn't happen years ago.

Unfortunately, I'm not fluent enough in a second language to make any sort of judgement call on AI versus Google Translate

At our company we have English, Spanish and Vietnamese. We never really liked the Google translate or bing results much. But we tested out the bing chat/gpt4 for translation and it did a phenomenal job. I asked multiple staff members their opinions on the translation and how much fixing needed to be done, not letting them know who did the translation. These are long documents, not brief phrases. I also showed them old translations by a previous staff member and asked them which they thought was better. I expected a broad mix of opinions but shockingly they thought the bing chat ai translation for Vietnamese was the best and didn't need editing (or very little). And these were native Vietnamese speakers where many of them were born in vietnam. They were shocked when I said it was AI. The Spanish translations did very well too though a couple people said it was fine but the phrasing could be better. But to be fair, it's hard for me to judge as our staff are mexican-americans born in the US with no formal Spanish training (they learned to speak it at home rather than school).

 

[onlysublime]

Yes. 'AI' (LLMs) will put together coherent, grammatically correct sentences (sometimes better than an average native speaker) that aren't a pain to read.

But as with all LLMs at the moment, they have a pretty big chance of those coherent sentences starting to go completely off-topic and bizarre.

I can only speak for Japanese, but LLMs can certainly put out something pretty decent that could just about be used as is. They are better than translation tools which within a sentence or two will have mistakes.

But as I said, you need to check if the LLMs are giving you grammatically correct gibberish, or worse. Translation tools stick to what you input, even if they sound odd or are using incorrect grammar.

Agreed but that's not the argument here. We aren't asking AI to answer questions. In this case we're only using AI for language translation and so far it's outperformed things like Google translate significantly.

 

[Findecanor]

Any site switching to AI for anything is already destined for the trash bin at this point. CNET, Gizmodo, et all have been on the decline for years and this is their overlords trying to milk the last pennies from the brand before it dies.

Indeed. This is a clear signal from these sites' management that even they don't have confidence in their own respective site any more, not even to keep up appearances.

BTW. I too was once tasked to help translate a web site to Spanish, not knowing the language. We used mostly Google Translate, but thankfully we had one person on the team who was a native Spanish speaker and good at communication. We relied on her a great deal to adjust the translation right.

 

[Honeybog]

At our company we have English, Spanish and Vietnamese. We never really liked the Google translate or bing results much. But we tested out the bing chat/gpt4 for translation and it did a phenomenal job. I asked multiple staff members their opinions on the translation and how much fixing needed to be done, not letting them know who did the translation. These are long documents, not brief phrases. I also showed them old translations by a previous staff member and asked them which they thought was better. I expected a broad mix of opinions but shockingly they thought the bing chat ai translation for Vietnamese was the best and didn't need editing (or very little). And these were native Vietnamese speakers where many of them were born in vietnam. They were shocked when I said it was AI. The Spanish translations did very well too though a couple people said it was fine but the phrasing could be better. But to be fair, it's hard for me to judge as our staff are mexican-americans born in the US with no formal Spanish training (they learned to speak it at home rather than school).

I’m not sure what your company’s audience is, but American Spanish is one of the hardest Spanish variants to get right. Your staff probably took exception with it because the AI was almost certainly trained on South American Spanish (all of the translation companies use South Americans, because the costs are so low) and reviewed by people with formal Spanish educations (e.g., Spain Spanish).

Don’t sell your Mexican American staff short. Assuming your audience is in the U.S., their editing is worth its weight in gold.

 

[GlockenspielHero]

Gizmodo articles have text? I just thought they were a pile of ads, covered up by auto playing video ads followed by a long list of Taboola article spam links

The various G/O sites are literally unusable on a phone- I can only read them on a desktop with multiple ad killers, and even then the comments don't work half the time.

 

[onlysublime]

Oh and one more observation... if I ask Google translate to do a translation, the result will pretty much always be the same. Not so with ai which has far more variance

 

[Honeybog]

even then the comments don't work half the time.

You say that like it’s a bad thing.

 

[earlyberd]

Companies firing their human workers because they think AI can adequately replace them are waaaaay too confident in the abilities of AI. As a bilingual person I constantly find errors in AI translations, so I know you still need a human to act as final editor to 99.99% of these translations.

Some languages like English are very logorrheic and make it very easy to glean the context from a passage. Other languages can have smaller vocabularies and more context sensitive, so a lot can be lost in translation that needs to be inferred by a human interpreter. And even those are prone to error.

These are the world's greatest tools for assisting a human interpreter in accomplishing their daily duties. They don't replace a human entirely, nor should they! Languages are constantly evolving and someone has to keep these models up to date with the latest lingo.

 

[GlockenspielHero]

A note to the bosses at Conde Nast. I pay money to subscribe to Ars because it has interesting articles written by actual humans who are very good at their jobs.

The day you decide you can replace them with an AI is the day I cancel that subscription. AT is worthless without them.

 

[onlysublime]

I’m not sure what your company’s audience is, but American Spanish is one of the hardest Spanish variants to get right. Your staff probably took exception with it because the AI was almost certainly trained on South American Spanish (all of the translation companies use South Americans, because the costs are so low) and reviewed by people with formal Spanish educations (e.g., Spain Spanish).

Don’t sell your Mexican American staff short. Assuming your audience is in the U.S., their editing is worth its weight in gold.

I'm not dismissing them but formal education does make a difference when it comes to writing and speaking. I'm sure grandma and others can be fine teachers but if they didn't have a rigored methodology to their trainings, there is a difference.

Case in point. We have a 23 year old who was born in Vietnam. She left for the US before high school. She looked at some translations by college-educated Vietnamese staff and made some recommendations. Our colleagues thought her translations were way too colloquial and crass. It's not that her translations didn't have their place but they were considered inappropriate for the source material (in this case, it was a flu vaccination advertisement).

And it wasn't like our Spanish-speaking staff didn't like the translations. Out of 10 people a couple had issues with phrasing.

 

[DrCreosote]

When I imagine AI fully taking over, I'm not half as worried about humanity as I am when I imagine AI being fully operated for the sole benefit of a handful of already rich ass holes willing to throw it all into the fire just to be more rich. We'll never be the benefactors of a cold and calculating thinking machine making choices for expedience alone. We'll instead be the benefactors of right bastards on a quest for more money.

 

[qazwart]

Por fin, una excusa para despedir gente.

AI English translation: “My hovercraft is full of eels”

 

[Honeybog]

I'm not dismissing them but formal education does make a difference when it comes to writing and speaking. I'm sure grandma and others can be fine teachers but if they didn't have a rigored methodology to their trainings, there is a difference.

Case in point. We have a 23 year old who was born in Vietnam. She left for the US before high school. She looked at some translations by college-educated Vietnamese staff and made some recommendations. Our colleagues thought her translations were way too colloquial and crass. It's not that her translations didn't have their place but they were considered inappropriate for the source material (in this case, it was a flu vaccination advertisement).

And it wasn't like our Spanish-speaking staff didn't like the translations. Out of 10 people a couple had issues with phrasing.

I would argue that formal education is entirely different from variant, though. I’m guessing your audience is either in California or Texas, since Vietnamese isn’t as important of a language on the East Coast. I routinely worked on vaccination campaigns where we actually wanted colloquial Spanish, and we could spot the fly-by-night translation companies a mile away, because they would always use South American or Spain Spanish.

 

[NomadUK]

— Tom Humberstone, The Nib.

 

[purposelycryptic]

Google Translate went online 17 years ago -- are the relatively new AIs really better at translating than Google with almost two decades of practice? If not, then I guess I'm curious why such transitions didn't happen years ago.

Unfortunately, I'm not fluent enough in a second language to make any sort of judgement call on AI versus Google Translate

AI should definitely be better, if only because it should have the ability to carry context from previous sentences into the translation of the current one.

I'm pretty sure Google Translate already uses systems similar to today's definition of AI for its output, since translations of a sentence can change depending on whether it is translated by itself or as part of a paragraph. The Google Translate of today is a far cry from even the version from a few years back.

I used to machine translate Japanese games and visual novels into Romaji and English as my base for creating a proper translation (since my Kanji knowledge is terribly lacking), and I frequently had to use several different engines just to get an idea of what was being said, as even just moving written Japanese to Romaji requires some context to get it right, despite them equating to the same spoken words.

It has definitely gotten a lot better now, with many translations immediately being able to bring the message across, even if they have to be rewritten to match style and characterization.

 

[Crash Bandicoor]

Ay Caramba!

 

[Wheels Of Confusion]

There's always the talking point that few people covered automation so worriedly when it was replacing auto workers. But those welding and painting robots generally took over tasks that were physically dangerous, mind-numbingly repetitive, and that required precision at high speeds. Those are exactly the kinds of jobs we should automate, for safety and sanity's sake. Especially if the end user's safety also literally rides on the line.

 

[YetAnotherPoster]

As I spent the first half of my life in the UK and the second half in Spain (so far, I'm not dead yet), I'm about as fluent in Spanish as I am in English.

IMHO Gawker's machine translation to Spanish is unreadable rubbish that makes your eyes bleed. In this example, "Doctor Who" was translated as "Médico Qué".

There's talk on social media of boycotting Gizmodo Spain over this, but there's no need for a boycot as people will just stop coming back anyway as it's unintelligible. It's like clicking on the wrong Google Search result and getting sent to some SEO page with random phrases and adverts splattered all over it.

And I guess that's why Gawker didn't shut down Gizmodo Spain - because they decided it would be more profitable to turn it into a landing page which serves adverts to the unwary. The ultimate in enshittified content for our times.

 

[Sajuuk]

This is all anti-human capitalism at its most blatant. But it's especially galling because sites like Gizmodo and CNet should, by virtue of "being" tech-oriented, be in every position to know beforehand that this shit is not going to work yet anyway.

C-suites literally can't even wait until the bots are ready before they start ruining their employees' livelihoods. I say chuck 'em into the incinerators.

I think it bears repeating in every thread like this: workers and owners are in opposition to one another, categorically. You are not a friend. You are not family. No, you are an expense in a spreadsheet they don't want to pay for.

 

[ArsPropaganda]

Any site switching to AI for anything is already destined for the trash bin at this point. CNET, Gizmodo, et all have been on the decline for years and this is their overlords trying to milk the last pennies from the brand before it dies. Sure newspapers like Gannet have also tried in the face of their own continued declicne but they seem to also recognize that the problems with AI do more damage than their worth and roll back on it (at least for the time being)

Three AI accounts downvoted this.