AI Reporter Bites Dog

January 23, 2023 6:22 PM

“In short, a close examination of the work produced by CNET's AI makes it seem less like a sophisticated text generator and more like an automated plagiarism machine, casually pumping out pilfered work that would get a human journalist fired.”
posted by gauche (33 comments total)

And a follow-up:
SEO Spammers Are Absolutely Thrilled Google Isn't Cracking Down on CNET's AI-Generated Articles - "Time to pump out content at ultra-high speed."
posted by thatwhichfalls at 6:30 PM on January 23, 2023 [5 favorites]

Yeah, we were just talking about this at work (another online tech publication). The way this type of AI works is it wants to make something that will be mistaken for a real article. It has no reason to produce "original" work, its entire existence and success depends on imitation.

I've known plenty of folks that worked over there. CNET is a big org and they just went through a rebrand. This is a really bad hit to their reputation when tech journalism is already hurting. Here's hoping instead of layoffs they take a serious stance against this sort of thing and set an example. I'm guessing editorial staff are fuming.

Google and others will catch up, but this is definitely a new arms race. I expected the bottom feeders like scrapers and squatters to use AI to generate low quality content but to have a big company like CNET do this is really harmful.
posted by BlackLeotardFront at 6:43 PM on January 23, 2023 [20 favorites]

Automatic fistulae pneumaticae
posted by clavdivs at 7:06 PM on January 23, 2023

"AI" generated text only sounds plausible to people because there is an enormous amount of text written by people fed into it. If you fed AI generated text into AI as training data over and over again, eventually the output would be bizarrely disconnected from any meaningful reality (and wouldn't resemble an article anyone would want to read). It cannot exist and function without us feeding it.

Which makes this an example of CNET eating someone else's seed corn. Garbage behavior really and can't stop soon enough.
posted by abucci at 7:08 PM on January 23, 2023 [15 favorites]

Doesn't matter if anyone reads it. They just need to click on it.
posted by thatwhichfalls at 7:10 PM on January 23, 2023 [4 favorites]

Do the infinite numbers of monkeys bother to stop and read what they are typing? Or what the monkeys on either side are typing?
posted by njohnson23 at 7:22 PM on January 23, 2023 [2 favorites]

Considering the malware issue with CNET some years ago I’m not likely to click on anything there. :/
posted by curious nu at 7:41 PM on January 23, 2023 [3 favorites]

What's particularly sad about text-generating AIs is that writing is not a task that humans are particularly averse to doing - there is certainly an adequate supply of people who will write copy for very little money. The promise of AI is text that you don't have to pay for at all, because the software steals it from someone else.
posted by anhedonic at 7:43 PM on January 23, 2023 [23 favorites]

SEO Spammers Are Absolutely Thrilled Google Isn't Cracking Down on CNET's AI-Generated Articles

So, replace AskMeFi with AskAI. Site funding solved!
posted by ChurchHatesTucker at 7:48 PM on January 23, 2023 [2 favorites]

Welp, so much for ChatGPT and whatever else taking over for all writers for free.... I hope.
posted by jenfullmoon at 7:50 PM on January 23, 2023

What's particularly sad about text-generating AIs is that writing is not a task that humans are particularly averse to doing - there is certainly an adequate supply of people who will write copy for very little money.

You could argue there's a lack of people who do it well, Sturgeon's law and all that, but I doubt this will solve that problem.
posted by Joakim Ziegler at 8:01 PM on January 23, 2023 [1 favorite]

Ultimately without a new web of trust or some other way to avert the infopocalypse, we'll need to revise Sturgeon's Law to 99% or higher.
posted by grokus at 8:06 PM on January 23, 2023 [1 favorite]

"AI" generated text only sounds plausible to people because there is an enormous amount of text written by people fed into it. If you fed AI generated text into AI as training data over and over again, eventually the output would be bizarrely disconnected from any meaningful reality (and wouldn't resemble an article anyone would want to read).

basically the 'grey goo' scenario for the web.
posted by logicpunk at 8:25 PM on January 23, 2023 [10 favorites]

I think I'd prefer a universe of interlocked paperclips to a wordcloud of marketing twaddle
posted by snuffleupagus at 8:29 PM on January 23, 2023 [7 favorites]

eventually the output would be bizarrely disconnected from any meaningful reality

And then we will achieve the dream - the AI CEO.
posted by His thoughts were red thoughts at 8:48 PM on January 23, 2023 [23 favorites]

I can't believe that there isn't at least one of us here playing fancy and feast with an all ai chat comment user.

Or we could just make an ai jonmc.
posted by NoThisIsPatrick at 10:29 PM on January 23, 2023 [5 favorites]

Why do I feel that "AI" writing only appeals to the sorts of people that never had any problem writing an essay that was 110% bullshit? Sounds like CNET will only publish good articles the same way I will visit them; by accident.
posted by krisjohn at 10:42 PM on January 23, 2023 [6 favorites]

Or we could just make an ai jonmc.

SWEATY GOLDBLUM FACE
posted by snuffleupagus at 12:08 AM on January 24, 2023 [4 favorites]

And then we will achieve the dream - the AI CEO.

Who do you think is ordering the layoffs? IT'S SKYNET!
posted by srboisvert at 1:55 AM on January 24, 2023 [3 favorites]

As ever, no 'I' present.
posted by GallonOfAlan at 4:44 AM on January 24, 2023 [1 favorite]

What's particularly sad about text-generating AIs is that writing is not a task that humans are particularly averse to doing - there is certainly an adequate supply of people who will write copy for very little money.

And the internet is also already filled with complete sludge written by humans who can’t really write, probably under-compensated and definitely unedited. This article is one of the funniest things I’ve seen this year - I’d love to make an FPP out of it, but I’m pretty sure it would break the guidelines for promoting illegal activities. It’s exceptional SEO filler crap for all the wrong reasons. But I can’t imagine it’s anything but IDGAF paid-by-the-word human-sourced. At least artificial intelligence has a working baseline of intelligence.

This AI stuff stands out far more for plagiarism than for any sins in its writing.
posted by Mchelly at 4:57 AM on January 24, 2023 [3 favorites]

It's getting worse. They're using ChatGPT to write legislation now.
posted by CheeseDigestsAll at 6:08 AM on January 24, 2023

The level of rewriting in most of these examples (bar one or two where there is indeed near-verbatim replication) would absolutely pass muster at any university I am familiar with, at both undegraduate and postgraduate level. (I used to be a thesis proofreader, my husband works in academia and does a lot of marking, as do several of my friends.) Sure, you'd be expected to cite your sources, but there is a much greater expectation of that kind of thing in academic work over journalism anyway. You wouldn't be expected to present it as a quote. A single reference at the end of a paragraph of several statements like that would more than suffice, but again, inline references are hardly a norm outside of academia.

I'm not saying there isn't a problem here, but I am saying there is more than a little hyperbole in how it is being described.
posted by Dysk at 6:17 AM on January 24, 2023 [1 favorite]

basically the 'grey goo' scenario for the web.

We've been through this with machine translation already, maybe close to a decade ago now. IIRC Google made some tweaks so that it wasn't so easy to have language-versions of websites that were just live-translated with GT. (You can still find plenty of grey goo if you poke around on Linguee, though.) I have the impression that subsequent changes to GT have made it less dependent on training data, which could be the way things eventually go with ChatGPT as well.
posted by Not A Thing at 6:39 AM on January 24, 2023

They're using ChatGPT to write legislation now.

This is perhaps the first positive use of this technology I've seen:

On Friday, state Sen. Barry Finegold (D) introduced legislation to set data privacy and security safeguards for the service and others like it that was “drafted with the help of ChatGPT.”
[....]
ChatGPT created a draft, later refined and formatted by Finegold’s office, that outlined restrictions against discriminatory data use and plagiarism and requirements that companies maintain “reasonable security practices,” according to screenshots shared with The Technology 202.

OTOH, when Big Tech discovers the level of influence that can come from automated legislation drafting, it's not going to be great.
posted by Not A Thing at 6:45 AM on January 24, 2023

I just.

What did those gullible fsckers who greenlit this at CNET think was gonna happen? Did they expend even one scintilla of effort understanding how text generators work? Did they even once glance at AI Weirdness?

I think ChatGPT is passing the Turing Test because this is how low the bar is. Every single dipshit at CNET who thought this was a good idea should be laid off immediately... ideally from the human race, but I'll take laid off from CNET and unhirable at any other nominally journalistic organization.
posted by humbug at 7:17 AM on January 24, 2023 [5 favorites]

We've been through this with machine translation already, maybe close to a decade ago now.

There's been translators on Twitter talking about how Google Translate fucked up their careers because businesses are happy with GT's halfassed machine translation and a lot of former clients are either simply replacing translators with Google outright, errors and imprecision be damned, or offering very small amounts of money to "clean up" after GT. Real shambles.
posted by Pope Guilty at 9:28 AM on January 24, 2023 [9 favorites]

This seems like it was inevitable; like it was designed to plagiarize from the get-go. Which, essentially, it was.
posted by tommasz at 10:49 AM on January 24, 2023 [3 favorites]

basically the 'grey goo' scenario for the web.

These new AIs are just getting good at it -- for years now I've been encountering doing a search for some computer or programming problem, and have to weed through pages and pages of programmatically reassembled content which may or may not have any relevance but they're so SEO-focused they show up all over the first few pages of every search engine. Or, the ones which have an answer way down at the bottom, but the first twenty paragraphs are suspiciously calculated paragraphs explaining minutae of definitions of words and themes that the answer is related to (like recipe blogs, but stupider), frequently repeating paragraphs with the same information but just rearranged in a different way. Maybe they're coming out of human-operated keyboards focused on meeting wordcount quotas but they're so devoid of content yet so focused on being seen, it's sort of a reverse Turing when you can't tell apart crappy content from AI content.
posted by AzraelBrown at 11:04 AM on January 24, 2023 [6 favorites]

So, replace AskMeFi with AskAI. Site funding solved!

Look, dont spoil this year's April 1 project.
posted by pwnguin at 1:16 PM on January 24, 2023 [6 favorites]

They're using ChatGPT to write legislation now.

If you've ever met legislators... well, this can only be an improvement...

OTOH, when Big Tech discovers the level of influence that can come from automated legislation drafting, it's not going to be great.

...until this.
posted by His thoughts were red thoughts at 3:41 PM on January 24, 2023

Every day we get closer to the Kafkaesque conclusion of humanity's foray into advanced technology.
posted by petiteviolette at 4:09 PM on January 24, 2023

Why do I feel that "AI" writing only appeals to the sorts of people that never had any problem writing an essay that was 110% bullshit?

"AI" writing also appeals to those of us who can't produce pages and pages of 110% bullshit but we're asked to do it. I'm definitely planning to use this for the next corporate "vision and strategy" document.
posted by gakiko at 5:36 AM on January 25, 2023 [2 favorites]