Barely a month after placing an indefinite pause on publishing AI-generated content, CNET owner Red Ventures appears poised to spin the AI content machine back up once again.
Leaked audio obtained by Futurism has revealed that the publishing giant's execs are gearing up for a second rollout of the AI tech, which Red Ventures has internally rebranded as an assistive "tooling" program. And this time, they say, things will be different.
"As you all recall, we put a temporary pause on AI-assisted content," said Lance Davis, Vice President of Content for Financial Services for CNET as well as its sister sites Bankrate and CreditCards.com, adding that the "first part" of a content audit — basically, combing its AI-generated articles for mistakes, a move made necessary by the revelation that the company's mystery machine was publishing blatantly false and sometimes plagiarized material — across "CNET, Bankrate, and Creditcards.com" has been completed.
"I want to talk about a few of the process changes we're planning to make, both to the tooling," added Davis, "as well to some of our editorial checkpoints that come up in the content creation process."
Davis then passed the mic to Red Ventures' Managing Editor of AI-Powered Content Cameron Hurta, who detailed "enhancements" being made to the mystery machine.
"Our primary tooling enhancements kind of go into three different buckets," said Hurta, listing them as "output," "quality controls," and "user experience."
As far as the output bit goes, Hurta promised that the company has replaced its "plagiarism-checking service" with something more "accurate" — a response to the revelation that much of the AI's work published before the pause had seemingly been closely cribbed from existing work by human writers, an issue that CNET's editors characterized as "phrases that were not entirely original."
Elsewhere, in regard to "quality control," Hurta promised that the new-and-improved machinery would offer "quality and AI likeness scores for all generations"— but not necessarily for the sake of the company's many millions of readers. Perhaps unsurprisingly, the execs seem primarily concerned about whether Google's algorithms will identify the content as AI-generated.
The likeness scores will "help us ensure that our content still reads human-like," said Hurta," and is not going to get tagged as being AI content."
"And that, you know," he added, seemingly as an afterthought, "the generations you're getting are of solid quality."
Given Hurta's concern about whether Google will be able to detect AI-generated content, it's not clear whether the Red Ventures sites will be marking AI-generated content as such in the future.
Asked whether AI content would be marked going forward, a Red Ventures spokesperson gave a statement that didn't seem to answer the question.
"CNET Money always identified the stories that were created with AI assist with a disclosure," they wrote. "We have made our disclosures even more prominent (as mentioned in CNET's statement here) as we continue to evolve editorial use of new technologies and tools."
Hurta also claimed that the new "tooling" system would integrate a few added editor-assisting features, notably in the form of adding source URLs into generated content. But if the saga of OpenAI's ChatGPT has proven anything, it's that incorporating fact-checking or source-finding services into language-generating systems is quite difficult. If the folks behind the Red Ventures AI content have figured out how to do so reliably, we'd be impressed. But considering that the likes of Google, Microsoft, and OpenAI haven't been able to do that quite yet, we're not entirely optimistic.
Circling back to those "editorial checkpoints," Davis promised that the company was assembling an "expert review board" that would fact-check all AI-generated articles, a requirement that would apply to CNET as well as Bankrate and Creditcards.com. That said, Red Ventures had previously promised that editors with relevant knowledge had reviewed its AI-generated articles — and that was before we reported on just how many errors were actually in CNET's financial explainers.
(And to that note, Davis also said in the meeting that Red Ventures had gotten an expert review on "most" of Bankrate's AI articles, "but not all of them.")
Here's hoping that the experts do a better job this time around, but forgive us for feeling a little gaslit by this particular pledge.
Davis also said that he would personally be carrying out a final review of AI-generated articles before they were published. Asked whether it was typical for a vice president at Red Ventures to be reviewing editorial content before it was published, a spokesperson said his title was "Editor in Charge of Editorial Content for CNET Money, Bankrate and CreditCards.com."
"He is responsible for overseeing and reviewing editorial content and is a member of the newsroom," the spokesperson said.
Neither Davis's LinkedIn nor Twitter lists that title, however, identifying him instead as a vice president at Red Ventures. Bankrate and CreditCards.com both list him as a vice president, and he doesn't have a bio at CNET at all — which seems unusual for an editor and "member of the newsroom."
And at the end of the day, regardless of the more existential debates surrounding AI journalism, the apparent Red Ventures push to get their AI back up and running just feels hasty. The public condemnation of the company's AI saga has been swift and embarrassing, and also, it wasn't that long ago. If just a month or so ago the system couldn't correctly calculate basic financial math, it's hard to imagine that the so-called "enhancements" have really covered that much ground.
But speed is apparently the name of the game. And unfortunately, it appears Red Ventures is sticking with its experiment-first, safety-later approach.
More on CNET: Leaked Messages Show How CNET's Parent Company Really Sees AI-Generated Content