Comment by jstarfish | Hacker News Reader

Comment by jstarfish | Hacker News Reader

Parent Full thread

> But Gannett insists the articles weren’t AI-generated. In a statement to The Post, a spokesperson said the articles — many of which have now been deleted — were created through a deal with a marketing firm to generate paid search-engine traffic. While Gannett concedes the original articles “did not meet our affiliate standards,” officials deny they were written by AI.

> Meanwhile, AdVon Commerce is open about its use of AI. On its LinkedIn page, the company says it uses “AI solutions for E Commerce.”

> A Gannett spokesperson took issue with the Guild claim that the writers weren’t real people, pointing to the LinkedIn page of one AdVon Commerce writer whose name appeared on a Reviewed article. At the top of his account, that writer touted his experience in “polishing AI generative text.”

> This isn’t the first time Gannett has found itself answering embarrassing questions about AI practices. In August, Gannett attempted to automate high school sports coverage at some of its papers with AI-generated articles, only to pull the experiment after it produced error-ridden stories.

It's like we can't trust mass media anymore.

There are two reasons to be disappointed, both because the articles were AI generated almost certainly (or lazy intern generated?) and because they were really paid advertisements and not identified as such. It's great that you gave those details, I just wanted to emphasize.

At my last company we had a dual use for the term AI.

When we anticipated that a function would eventually be automated by artificial intelligence, we would designate as a task force AI.

Meanwhile, the AI would start out as “actually, interns” until we built up enough operational data to create a training suite. So in our case, AI often meant “Actually Interns”.

Sometimes we contracted a firm in India for mechanical Turk type tasks as well, so it also worked for that.

When you thought something could be automated by ai, were you thinking soon? Like say adjusting memory and gc settings for java in a running app? I just don't see ai solving problems that I run into like that in the next few years. I do see ai helping people write much more code maybe, work in domains that they don't understand well (I want to add new php functions to do blah to extend my company's website, but i don't know anything about php).

So write me prototype code to do something that I can describe in brief, but skeptical of doing things that are futzing with per-application use case cloud settings. I don't want to have to learn how you adjust foo or bar or baz on google cloud and aws and azure and ... - automating that is useful. I'd like to automate say java memory changes, but making the gc system do it without me needing to change flags is even better.

Usually it was tasks involving “human” judgement that did not lend themselves to conventional multivariate mappings or algorithms. When we transitioned our “proof of personhood” task to AI based on collected and curated training data, we achieved less false negatives and positives as well as a marked improvement in user satisfaction with the process.

Most users thought we gave up on AI and went to real people when we transitioned from humans to AI lol.

The main reason to be disappointed is the fluency with which people have taken to just outright lying. The article, to me, proves pretty conclusively that the articles were AI generated and that most (or at least some) of the authors aren't real people but the spokesman is just... lying.

> A Gannett spokesperson took issue with the Guild claim that the writers weren’t real people, pointing to the LinkedIn page of one AdVon Commerce writer whose name appeared on a Reviewed article. At the top of his account, that writer touted his experience in “polishing AI generative text.”

Like... This is like pointing to a robot and saying it isn't a robot. Nobody is fooled. But... why even do it? Isn't it... idk, embarrassing? To just talk complete nonsense?

> The main reason to be disappointed is the fluency with which people have taken to just outright lying. The article, to me, proves pretty conclusively that the articles were AI generated and that most (or at least some) of the authors aren't real people but the spokesman is just... lying.

Agreed. Maybe there's a proper term for it but I call it the "hitman fallacy," where people reason that they can't be accused of murder because they aren't the party that pulled the trigger.

In a way, I can see their [tortured] logic-- they did in fact pay real money to a real person to write that article. It should follow that a real person did in fact write it, otherwise it's fraud.

Gannett shies away from calling it that because it's not fraud when the other party delivers what you asked for.

> It’s like we can’t trust mass media anymore.

We can’t trust Gannett, but it’s dangerous to dismiss all perceived “mass media” because of the decisions of one company.

If you’ve been paying attention to who owns the media you consume, you’ve likely already written off Gannett-tarnished outlets years ago.

Reply on news.ycombinator.com