USA Today Owner Pauses AI Articles After Butchering Sports Coverage | Hacker News Reader

USA Today Owner Pauses AI Articles After Butchering Sports Coverage

futurism.com

50 pointsnxten27 comments

I don't know about you, but "a close encounter of the athletic kind" is a hit for me.

People think journalism is draped in ulterior motives now, just wait until these publishers are just front ends for AI content created by private companies.

What’s the point in having sponsored content when your supposedly unpaid content is churned out by some kids who raised a couple million bucks off of this AI wave?

The ultimate goal of advertisement is to be invisible, to be influencing without having to admit as much, and AI is the perfect vehicle to do just that.

To be fair, just because we can generate text well doesn't mean we're any closer to actually knowing how to influence people subtly like that. Anything that makes content can be used for vague general propaganda/lobbying, but impacting product-decision in a specific way? Seems as difficult as ever.

> Anything that makes content can be used for vague general propaganda/lobbying, but impacting product-decision in a specific way?

A/B testing and tracking metrics on user behavior is a very mature practice in the industry. Some types of cases and behaviors might be hard to track, but then really obvious ones (like Jan 6) also stand out.

High profile universities like Stanford built behavioral design labs backed by millions in funding to advance our understanding of how to influence behavior through technology. This is a massive commercial research space with huge money on the line. We can’t ignore that.

the owner is gannett.

the butchered coverage was in local newspapers, also owned by gannett.

which begs the question, why doesn’t the headline just, you know, name gannett?

edit: anyone downvoting care to indicate why? it seems rather pertinent to name the owner, given the scale, and the potential impact of using such technology at scale.

I generally downvote complaints about the title of articles. (I don't downvote requests to change the title on HN though.)

there was a bit more to my comment than a “this title is bad.”

do you know of gannett? legitimately curious.

The thesis of your comment seems to be that Gannet should have been in the title, with your other two points being supporting evidence.

I've never heard of Gannet before today.

I don't really mean much by the downvote. It's a personal pet peeve of mine when people comment on the headline, and I think it's against the spirit of this part of the guidelines:

> Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting.

My downvote definitely didn't mean that I didn't think there was something substantive to say about Gannet's involvement, or that I disagreed.

(The guidelines also tell us not to talk about voting, which I'm breaking of course, but I do try to respond when people genuinely want to know why I voted a certain way. I understand why that part of the guidelines exist, but I have mixed feelings about it, I don't see how else people are meant to know why they were downvoted.)

> I've never heard of Gannet before today.

this is my very point, and it is anything but tangential.

thank you.

People think of Google, not Alphabet; of Facebook, not Meta. The headline identifies the brand familiar to millions.

the examples you list all involve the “smaller” company effectively becoming the larger.

gannett devours and typically avoids visibility. if nothing else because it’s confusing to the commoners that so many different news outlets are actually the same, without admitting it.

so, yes, what you said. but not quite.

"Gannets" is how some people describe hungry seabirds that eat everything, and by extension children (by my mother) and meeting attendees (by our office admins) who demolish food offerings. So the company Gannett is quite well-named!

Where does the data on the sports event come from? There must be other articles already written about it for the AI to attempt to reproduce, right?

Why wouldn’t they just use those original articles? Or are these made up whole cloth and a CEO thought AI just magically knows what happens on high school football fields?

They start with raw data collected at the event itself. A lot of games have statisticians who collect not just the scores but gather data on every play/at-bat/etc. Even high school games have those; it's often the official job of some student.

That does have enough information to make an article. In the early days of baseball, a radio announcer could receive the stats continuously and narrate the game, with a bit of imagination.

Some of the examples look as if they got nothing more than the box scores, and used AI hallucination on other sports articles to generate a bunch of generic verbiage out of that. (To me, as a non sports fan, that's what most sports writing looks like anyway. I have no idea why people want to read that, but there is a market for it.)

I'm surprised it failed this hard. AI writing has been used for a while in fantasy sports platforms for a while to put in little quips about player performances based on their recent stats and automated projections and read pretty well.

Probably a lazy implementation that they farmed out to the lowest bidder

AI for content creation is still in its infancy. It will get better, smarter, and more effective. Even humans make mistakes sometimes.

We’re not at a point where this can be a human free pipeline, but I bet you could make this work with 1-2 humans in the pipeline that review instead of the 100 you would traditionally need.

In theory, but I think the problem you'd find in practice is that reviewing 100s of nearly-identical AI-written articles gets very boring very quickly, and a lot of errors would slip through.

Ok, so 2 is not enough. What if you put 4 humans? Or 8? Or maybe even 20 (instead of 100, per OP)?

These experiments fail because they greedily try to outsource all the work to AI. Another recent example: the lawyer who submitted ChapGPT hallucinations directly to a court case.

You don't need to eliminate humans, and certainly not at first. You just need to be much more efficient than status quo in order for AI to be deployed at scale.

Why? Lots of people do a lot of reading for a living. Reading a new article is way more interesting than thousands of jobs I can think of. Data analysts and accountants literally pour though millions of featureless numbers over their career, why would this be any different?

It's like trying to find typos in your own writing. It's very difficult to stay focused when reading a long series of nearly identical documents. The people you're talking about are reading a lot of novel documents.

Why would these be identical? New stories everyday, new games played etc.

Because they're populating a template.

> One article even failed to populate properly, with the text instead featuring a bracketed glimpse at how its opening sentence was supposed to read.

> "The Worthington Christian [[WINNING_TEAM_MASCOT]] defeated the Westerville North [[LOSING_TEAM_MASCOT]] 2-1 in an Ohio boys soccer game on Saturday," reads the butchered intro.

> Why would these be identical? New stories everyday, new games played etc.

If the only data you have about a game is the sport, the team names, and the score, there's only so much an AI (or a human!) can do to write an interesting article about it. Once you've read a few dozen of the generated articles, they'll all start sounding the same -- because, aside from the details, they are all the same.

If you want quality articles, you need some more depth in the source data. And, for little-league sports games, that data may just not exist.

> Data analysts and accountants literally pour though millions of featureless numbers over their career, why would this be any different?

The numbers are probably meaningful to them due to their education though.

It’s like saying programmers spend their whole career pour though millions of lines of nigh random symbols they call source code.

Producing something error free sure. Producing something beyond clickbait and spam that people actually want to read, I seriously doubt it.

Reply on news.ycombinator.com