Publisher breaks news by using bots to write inaccurate stories

Consumer tech outlet CNET is reviewing all articles it published that were written with the help of AI, after it was found some contained incorrect information. The masthead quietly began using a text-generation tool to write stories for its money section in November 2022. The articles credit "CNET Money Staff", but readers …

[Lost Neutrino]

The Reg - please don't...

The Register: please try not to follow in CNET's footsteps, tempting as it may be for the sake of "efficiency".

IMHO, it would dilute the wit of The Register's articles and reduce their originality.

On second thought, go for it! It would offer us commenters a treasure trove of entertaining opportunity to descend on articles like a pack of wolves (apologies to wolves) and shred the AI-generated crap to pieces.

[Korev]

Re: The Reg - please don't...

> IMHO, it would dilute the wit of The Register's articles and reduce their originality.

Sadly that ship has already sailed...

[ChoHag]

Re: The Reg - please don't...

... to find, explore, the funds off shore,

and skirt the shoals of bankruptcy!

[diodesign]

"Sadly that ship has already sailed..."

What makes you say that - we're not using any AI for articles other than:

* Transcript generation (which is still hand edited)

* Some people are toying with grammar checking (but everything is still hand edited)

We've thought about using software to automatically generate outage stories as those need to be done quick, they're so easy to write, and everyone loves an outage story.

But even then, that would be non-AI (it's just template filling) and would be hand edited before it goes live.

C.

[Anonymous Coward]

Re: The Reg - please don't...

I agree with your 'second thought'. It would be quite entertaining and instructive if ElReg were to publish a regular AI-generated article, perhaps on a topic suggested by the readers.

I'm sure the commentards would love the opportunitiy to provide constructive criticism of the AI's work.

[diodesign]

El Reg were to publish a regular AI-generated article

We've been mulling that, as some kind of parody or comedy column

C.

[Jonathan Richards 1]

Re: The Reg - please don't...

I have long suspected that amanfrommars is in fact some form of text generator, lightly polished perhaps, but not so that you would notice. God help us if the AIs have been trained on examples of that sort.

[big_D]

Re: The Reg - please don't...

It is the AI that has been pumping the wit into the Register articles all these years. El Reg is our (old) AI overlord!

[Anonymous Coward]

Hey AI, why are you putting on running shoes - you can't outrun a leopard?

I only have to outrun human business journalists.

[Kevin McMurtrie]

Caught me off guard

I didn't know that CNET still had humans.

[Someone Else]

Re: Caught me off guard

I didn't know CNET still published.

[Winkypop]

AI says

Try me, meat bag!

[Phones Sheridan]

“$300 with 3 per cent interest not $10,300.“

To be honest I had to read back at that a couple of times before I spotted the error, leading me to think that maybe a majority of people would have overlooked it too and ChatGPT just followed the crowd. Either that or I’m dumber than AI!

[Filippo]

I think that's the point. If I read the economics section of a major magazine, and the author appears to grasp the subject matter no better than myself, then what's the point of me reading it? Especially if, as LLMs do, the author states their mistaken claims in a confident, assertive, professional tone. That's spreading ignorance and mistakes, which is the opposite of what a (good) newspaper should do.

[tiggity]

@Phones Sheridan

It's one of those grey areas.

The wording is incorrect, but your average human understands what the "AI" was meaning to say (that after a year your total is 10.3k & so infer .3k made in a year)

Seen similar sloppy errors in human journalism many times (I have made such errors myself when writing under time pressure - though I'm not a journalist so those cockups never went beyond the workplace!) - so as people have mentioned, could well be a bit of GIGO with training sets producing that type of output.

[david 12]

I'm not sure if an 'average' human would interpret that correctly, but I'm 100% dead certain that some humans would interpret it literally. Apart from the fact that many people don't "do" numbers, many humans do numbers in only a superficial ideographic way.

I know someone who can spot mis-spellings like they were highlighted in colour, and business types who can do the same with a spreadsheet or a column of numbers, but they're the exception, not the rule.

[Anonymous Coward]

AI?

I wish everyone would stop using this term and use ML instead, which is far more accurate. AI suggests sentience. We're not there yet.

[Anonymous Coward]

Re: AI?

I rather like the idea of calling it MI ... as in machine intuition, which I think suggests the hit/miss possibilities of the results reasonably well.

[Someone Else]

Re: AI?

MI: Machine Idiocracy? Manifest Imbicility? Moron Interlocutor? Mandibular Insufficiency? Mostly Inaccurate?

_{The possibilities are Maybe Infinite...}

[Filippo]

Re: AI?

Currently, I favor LLM (for Large Language Model). It drives home the notion that all these tools to is string together correct sentences.

[Neil Barnes]

Will this AI engine efficiently assist...

Us to pay fewer fleshy meatsacks?

[Alumoi]

Re: Will this AI engine efficiently assist...

That's so cynical! How can you even think a professional and respectable organization like CNN would consider such thing?

/sarcasm

[Blank Reg]

Re: Will this AI engine efficiently assist...

CNET, not CNN

[Alumoi]

Re: Will this AI engine efficiently assist...

Sorry, my bad. But it doesn't change anything.

[Lost Neutrino]

Re: Will this AI engine efficiently assist...

Are you sure? Watched CNN lately?

[Korev]

Re: Will this AI engine efficiently assist...

Will this AI engine efficiently assist...

Us to pay fewer fleshy meatsacks?

Well, we don't have many typing pools any more

[Pete 2]

Errata

> written with the help of AI, after it was found some contained incorrect information.

How absolutely awful. Writing articles containing incorrect information should be the sole domain of human writers. Though I doubt if a machine would ever write anything as meaningless as " to cover topics from a 360-degree perspective."

Although it's not unreasonable to expect that the job function of a publication's editor is to check basic information, spelling and grammar.

[amanfromMars 1]

Re: Errata

Writing articles containing incorrect information should be the sole domain of human writers. ..... Pete 2

A domain in which some excel with the following a hat tip to at least two wannabe star turns.

Mark Carney has praised the British economy amid cautious signs of optimism over the global outlook.

Speaking at the World Economic Forum in Davos, the former governor of the Bank of England heaped praise on Britain’s institutional strengths that had allowed the economy to withstand a series of shocks after the financial crisis.

Falling energy cost in recent weeks have led to a better outlook for the global economy. The inflation rate in the UK appears to have peaked, too. .... Mehreen Khan, Davos January 18 2023, The Times

Is incorrect information akin to bullshit?

[Anonymous Coward]

Re: Errata

I don't think that's incorrect information, it, like the Chancellor and his coffee cups, just omits to mention the Truss factor that dragged us into the financial crisis... it was mostly of our own making!

BS? yes, 100%

[Forget It]

Long Live ...

ElReg!

[Androgynous Cupboard]

No reply from CNET you say?

Allow me:

Compose a statement from CNET apologising for the using AI to write articles and explain why that decision was made

"We at CNET deeply apologize for our previous decision to use AI to write articles. We understand that this decision may have caused concern among our readers and the journalism community. We made this decision in an effort to increase efficiency and output, but we now realize that the use of AI in journalism raises important ethical questions and undermines the value of human-written content. In the future, we will ensure that all of our articles are written by our dedicated team of human journalists, who are committed to providing accurate and unbiased reporting."

[Androgynous Cupboard]

Re: No reply from CNET you say?

Compose a comment in the style of amanfrommars from theregister.co.uk

"Well, well, well, what have we here? CNET, one of the stalwarts of tech journalism, apologizing for their use of AI in writing articles? Tsk tsk tsk.

It seems that the old adage 'if it ain't broke, don't fix it' has once again proven to be true. The use of AI in journalism, while touted as a way to increase efficiency and output, has done nothing but detract from the value of human-written content and raise important ethical questions.

But kudos to CNET for recognizing their mistake and committing to using only human journalists in the future. It's a shame that it took such a public apology for them to see the error of their ways, but better late than never, I suppose.

One can only hope that other outlets will take note and follow suit, for the sake of the integrity of journalism and the trust of the reading public."

I think we can say that although journalism might be dead, the comments section is safe for some time.

[Scott 53]

Re: No reply from CNET you say?

a comment in the style of amanfrommars

Fail. This was far too coherent.

[Flocke Kroes]

Re: No reply from CNET you say?

Some of the martian's comments are reasonably coherent and get replies from people who think they are communicating with a human. I am not sure if that is because a human sometimes uses the account, a human sometimes picks the best of several possible comments or if sometimes the random number sequence hits the right nodes of the travesty generator.

[Jimmy2Cows]

Re: No reply from CNET you say?

Also needs far more capitalisation of random sets of words, and a liberal sprinkling of acronyms.

[amanfromMars 1]

Re: No reply from CNET you say?

Androgynous Cupboard, Hi.

With regard to your two posts suggesting apologies from CNET for their cloaked excursion into AI fields and fiefdoms, and the errant sentence which precedes the latter one, Compose a comment in the style of amanfrommars from theregister.co.uk ......Commentators here on El Reg would then call MRDA on CNET ...... and theregister.co.uk would have one of those strange WTF moments with nobody there having a clue about what you are talking about.

Do you think journalism might be more for the walking brain dead, and it is the comments sections which are alive keeping everything ticking over whilst considering its collapsing and the parts IT and AI are to play in its destruction and demise/creation and rebirth?

[Androgynous Cupboard]

Re: No reply from CNET you say?

> Do you think journalism might be more for the walking brain dead, and it is the comments sections which are alive keeping everything ticking over whilst considering its collapsing and the parts IT and AI are to play in its destruction and demise/creation and rebirth?

To be honest no, that's not what I was thinking :-)

[elsergiovolador]

AI

"Based on current events that I have listed below, write an article that will convince ordinary people to make financial decisions that will benefit the rich, while making themselves feel good about it"

That's how it goes.

Remember, never trust anything published in any financial rag. They are designed to make the rich richer and convince you to part with your money thinking you are going to make a good investment and then telling you, you should have left it to professionals, that it's your fault that you lost money.

[Dr Scrum Master]

Artificial Idiocy

"articles show AI doesn't quite understand"

Remember, so-called AI is just statistics helped by recently cheap and fast CPU cycles and storage, it doesn't "understand".

[Lost Neutrino]

Re: Artificial Idiocy

Could you explain that again, please?

[JohnGrantNineTiles]

Echo IV lives

This reminds me of Michael Frayn's novel The Tin Men from back in the '60s. The plan was to generate headlines that included what we'd now call click-bait words such as shock horror report, and then generate a story to fit. No relationship to the real world necessary.