Last week, Gizmodo owner G/O Media shut down the blog's Spanish-language site, Gizmodo en Español. The humans who worked at the blog, which ran original stories and adapted English-language Gizmodo stories into Spanish, were fired — and, ominously, were replaced with an AI-powered translation tool.
"Hello friends," former Gizmodo en Español writer Matías S. Zavia wrote in a post to Twitter-formerly-X. "On Tuesday they shut down @GizmodoES to turn it into a translation self-publisher (an AI took my job, literally)."
As noted last week in The Verge, which first reported the layoffs, Gizmodo en Español articles are now outfitted with an AI disclaimer explaining that each article's "contents have been automatically translated from the original," and that "due to the nuances of machine translation, there can be slight differences."
That's disheartening, and a striking example of an AI model directly replacing human roles. And to make the whole ordeal even worse? The AI isn't even doing a good job: as its disclaimer suggests, it's already making some glaring mistakes.
The model started to slip up right out of the gate. As a Gizmodo en Español reader pointed out last week, just a few short days after the site's culling, one of the AI articles didn't even translate all the way through, instead switching back to English partway through an article.
Hace unos días @GizmodoES despidió a sus redactores en español para pasar a solo publicar traducciones de su edición de USA vía inteligencia artificial.
El resultado: textos que de pronto te cambian de castellano a inglés y titulares que te dejan con el culo torcido. pic.twitter.com/Uumvvdi5AW
— Víctor Millán (@victorcmn) September 1, 2023
The mistakes didn't end there. When we combed through the site's AI-translated blogs, we found that the system is prone to gaffes like basic formatting errors and technological fails.
Take the translated version of a recent Star Wars recap, which in English was titled "'Ahsoka' Had Another Major Cameo You Definitely Missed." The AI seemingly doesn't quite understand how to populate the page, cutting off paragraphs at strange points and blundering punctuation. One paragraph even includes a bit of HTML that definitely wasn't supposed to make it onto the page, reading: "<caption>¡Kanán!</caption>"
A human editor surely would've prevented these mistakes from making it to print, and though punctuation mishaps happen to the best of us, a human writer presumably wouldn't have included random pieces of HTML in the first place.
Other AI-translated articles, like this one about STIs, experience similar formatting problems. The STI article also illustrates another formatting pitfall for the AI: outside links. The system is apparently very bad at properly repopulating any in-text links present in the original article, unnecessarily linking massive chunks of entire paragraphs when the initial write-up only attached a given link to a word or two.
The blunders continue throughout the many AI-worked blogs. One about the planet Jupiter features missed spacing and formatting bizarreness, in addition to some mysterious floating punctuation; this post about the events of this past "Barbenheimer" summer is similarly mangled; and yet another space article, this one about the ALMA telescope, falls to pieces at the end.
All of which, again, human staffers would almost certainly have caught before any of these pieces went to publish — a detail that doesn't just punctuate the AI's shortcomings, but also appears to suggest that G/O's new, automated system is lacking in any serious human oversight whatsoever.
When we reached out to G/O for comment we didn't receive any response, but we did notice that they've since made edits to at least one of the articles that we'd flagged over email.
It's worth noting that this isn't G/O's first foray into AI-generated content. The publisher came under fire last month for its decision to publish AI-generated material on its flagship sites including Gizmodo and The AV Club — much to the anger and frustration of those sites' human staff, who were warned of the publisher's new AI effort just minutes before the first AI-spun article was posted. As for the content itself? It was awful. Wildly incorrect, with glaring grammatical errors to boot.
That in mind, it's almost impressive that G/O went through with yet another AI effort so soon afterward, seemingly without bothering to soundly test the new AI for what seem to be foundational issues.
Instead, it appears that G/O has learned nothing from its mistakes, nevermind those at CNET and Bankrate . Or maybe it's just so keen on eliminating human roles in favor of automated content that it's willing to suffer the embarrassment. Either way, it's depressing.
More on Gizmodo and AI: Gizmodo and Kotaku Staff Furious After Owner Announces Move to AI Content
Share This Article