"The Associated Press created this story using technology provided by Data Skrive and data from Sportradar."
Example paragraph:
"The two teams match up for the second time this season. The Nuggets defeated the Clippers 111-108 in their last meeting on Nov. 15. Jokic led the Nuggets with 32 points, and Paul George led the Clippers with 35 points."
100% generated from the stats table, and totally boring and devoid of life. Horrible.
Yahoo fantasy football does this with each weekly head to head matchup. They're clearly labeled as machine generated ("Powered by ChatGPT API") and add a little fun.
The reason is people will click on them thinking it is an interesting story. The publisher (ESPN/Yahoo/etc.) just care that you click on the page so the ads load and they get the impressions. Some PM probably ran some analysis that machine generated articles from a stats table will get X number of clicks which will generate Y dollars in ad revenue. There was likely no consideration that the overall content of the site would decrease. After awhile people stop clicking on them because they know it will be a machine generated article so eventually the publisher will stop putting them on their website.
I think the industry term for this is "made for advertising content".
The PM or whosever's idea it was: "we don't do long-term experiments so nobody can prove it is ultimately destructive and imma be outa here next year anyway".
Will the Swimsuit Edition be generated with Stable Diffusion?
It certainly could be. Go here.[1] Use prompt "Sports Illustrated Swimsuit Edition cover." Under "Advanced", select model "ICantBelieveItsNotPhotography". Click Generate.
I mean, the contents of the Swimsuit Edition, like all magazine-ish pictures of women intended to generate That Kind Of Attention, have been far from organic for quite some time, so it's not that much of a difference.
FWIW, I don't think that one was even using any kind of modern AI. The bizarre phrasing is characteristic of the article having been run through a "spinner" tool which randomly substitutes words or phrases for synonyms, without any regard to context -- which is how it ended up saying he "handed away" (passed away) after having played as an "ahead" (forward). These tools have existed since the late 2000s, but were mostly used by SEO spammers.
It's hard to pin down exactly what I find so unsettling about the practice – it's almost like the uncanny valley, but for written content that apes human expression instead of imagery?
Might not be exactly this, but it makes me feel similar to why people hate advertising. Which I believe is, people don't like feeling lied-to, and everyone knows that marketers are trying to get in your head to manipulate you into manufacturing desire or stoking insecurity, all for the purposes of getting you to buy their products.
I think people like organic word-of-mouth, but on the flipside, hate when they find out that someone was a paid shill to posture as an average consumer, but are an industry plant to trick and deceive us all lol.
But to your point about why it feels icky and unsettling for publications & media companies to just straight-up use AI to write articles... seems kinda similar. Many of us are already skeptical that journalists & reporters are being censored and manipulated into writing with an agenda. But these types of AI-generated articles feels a few degrees more dehumanizing and Machiavellian. Like, the humanity aspect can all be aped so well, that we can just manipulate the masses and assuage their needs for a sense of connection without having any souls behind it whatsoever, because the masses are viewed as a bunch of manipulable "things" to simply extract things from (like attention).
I don't like it either, and for me it seems like it's those reasons. It feels so... gross and heartless.
One of the reasons I find so many ads annoying: the copy is complete shit. It's usually vapid, kitschy, cringy garbage. Most ads are like a Joss Wheadon show; formulaic, cookie-cutter "clever" that appeals to the simplest minds. Nobody talks like that in the real world.
It also usually feels like the creative process was supervised by a bunch of people who seem to think themselves a superior sort of human.
This type of behavior will only serve to cheapen content across the board. At this point, even the word "content" betrays the emptiness of it all - people don't pay for bags and boxes of "content" do they?
AI "content" is a nothing-burger. It is inherently devoid of "value" and seems like a last-ditch effort to squeegee the remaining drops of attention off of everyone's eyeballs without actually investing in genuine creativity.
As more and more of this dross floods the Internet, the very purpose of the web may be called into question. How can we share information with each other if the world's library/archive becomes the world's bot-poop landfill?
The Internet has evolved from a shared information system to so much more, so I hope this unfortunate phase will soon pass and ML tech can be put to more appropriate use than just crapping out low-effort "content" all over the place.
> At this point, even the word "content" betrays the emptiness of it all
In all fairness, "content" telegraphed that from the first time it was used in the online sense. I still don't understand why people are willing to use it to refer to their own work.
I'd be very, very curious to hear viable alternatives to ad supported models, that aren't based on how major companies have been doing it for the past 20 years (or longer?), where they make it free, then sneak in subscriptions, then over time, start increasing subscription costs.
I feel like it's not just the companies, but consumers/ audiences don't want to pay for most internet services unless it's something like infrastructure services where it somehow viscerally seems "sensible" and "right" to do so.
All those online services, even the AI ones, need servers and bandwidth and paid people to maintain them. They need sources of income and the only reliable one known to man (today) is ads. If you have a different idea, please do explain, but before you say "monthly contributions" think twice about the reality of it and maybe also count for how many services you personally pay. Of course one could say "why do I need online services at all" but that's not a future I care about.
The unsettling thing about it is that it's a lie from front to back, intended to deceive people into believing there are real people sharing recipes, when the people don't exist and nobody has ever eaten the food.
The opposite would be great: a web crawler that digests (pardon the pun) a 5mb web page (or 20m long-winded video) on how to cook a meal and condenses it into just the relevant steps and photos.
https://mela.recipes/ does this with its built-in browser/parser thing (and it has a bunch more incredibly useful stuff).
Highly recommended, it's from Silvio Rizzi of "Reeder" fame so it's a one-time purchase built with extreme care by a solo dev with excellent product instincts. Huge fan of his work, this kind of high-craftsmanship software is just so pleasant to use.
There is a whole slew of AI based video summarizing tools in existence already, enough so that a search for "video summarizer" has a bunch of listicles in the results.
The doomsday clock for the fun parts of the internet is reaching midnight. This is what generative AI will be unleashing in droves as the software becomes more mature and eliminates the giveaway sentences and phrasing.
This seems to depend heavily on what you consider 'fun'. I agree that AI is taking over these spaces and that it's alarming, but OTOH a lot of the examples in this thread (sports chatter, recipes etc) are mostly crap, because they are dominated by microcelebrities whose persona is intense obsession with one topic (and which persona is often just the brand of a sophisticated marketing operation).
This was a good meme a year ago but these days it’s out of date. Popular image AIs like Midjourney and Dall-E now do accurate hands the vast majority of the time.
Not really, they still can't do hands and faces reliably, especially DALL-E which hallucinates like mad (not just with hands) and generally has quite a lot of distinctive artifacts on lines and densely packed details. It is possible to fix hands in SD/SDXL 100% of the time, but that's much more involved.
The real 'winners' here in the upcoming AI wave will be people with existing platforms. They'll replace staff and pocket the difference. This one was obvious but it soon won't be.
> "After we reached out with questions to the magazine's publisher, The Arena Group, all the AI-generated authors disappeared from Sports Illustrated's site without explanation. Our questions received no response."
What is The Arena Group? (https://thearenagroup.net/). It's a publicly traded company for one. (Stock price: AREN (NYSEAMERICAN) $2.76 -0.03 (-1.08%))
> "The Arena Group is an innovative technology platform and media company with a proven cutting-edge playbook that transforms media brands. We aggregate content across a diverse portfolio of over 265 brands, reaching over 100 million users monthly."
So, basically, an entity that has people's eyeballs, content doesn't matter that much does it? But brand does (SI has notoriety for millions of people). I'm guessing ads are the main business here, therefore content generation in all ways that get people's attention is the goal (for cheap).
> The only problem? Outside of Sports Illustrated, Drew Ortiz doesn't seem to exist. He has no social media presence and no publishing history.
Looks like they're still working out the kinks. I look forward to the internet three years from now, where AI-generated authors have matching LinkedIn profiles and active social media accounts.
This battle will not be won by anyone other than the people using these tools to their advantage to pump and dump.
The genie is long out of the box now. Future iterations of LLMs will not get worse but better. And already now, something like GPT-4 easily bypasses human detection if the output is inherently controlled by a human.
Bad AI content can be detected super easily. ChatGPT is limited by its system prompts and it will always take the “least effort” way to answering your question, be it a question or an instruction to write an article. Repetition is a massive issue with 3.5 and Google can scout that out blindfolded.
If you want to mess with your own reputation then by all means use AI. The average internet user will not be any wiser about it. I would be very surprised if Google took action against these types of campaigns based on user feedback as opposed to an implementation in their own algorithms.
The internet has gone past the stage of enshittification.
Everyone knows, or intuits, that the game is up. This is the end game, the Shit Squeeze, where the last drops of goodness are wrung by force from what once was something exciting... and the flames are being fanned by generative AI.
The game has been up for a while, this just takes into hyper-drive.
But don't forget, you _can_ opt out of this corporate and consumerist side of the internet. It's over when you're convinced that you cannot do so.
For me personally, this has made it easier to step away from places on the internet that had already started to go down the drain. It's as if the shitty part of internet finally consumes itself out of existence.
Instead, I now follow small personal blogs and niche forums where this is not an issue. Just people posting because they still believe in sharing and connecting, despite of generative AI.
It's different in one way in particular: AI-generated content is in the public domain in the US, since copyright isn't applicable until a human's creative input occurs. That's either better or worse, depending on your viewpoint!
That'd be like saying all oil paintings are in the public domain because paintbrushes can't hold copyright. The copyright goes to the person triggering the generation of the content similar to how if you use blur tool in photoshop you don't suddenly lose copyright.
> Maybe that will be the case in future after a court decides so, but at present AI content does not get copyright protection.
That's not what the article says.
The article says that an AI cannot hold copyright. This is not the same as AI content cannot be copyrighted. As I said before, a paintbrush cannot hold copyright; that does not mean a painter who used a paintbrush can't hold the copyright to the work.
This is basically the same thing as when that guy left a camera out and PETA tried to get the monkey that took the photo to be assigned copyright [1]. PETA failed; animals cannot hold copyright and the guy that left his camera out can.
You're forgetting that the guy who left the camera out for the monkeys to take pictures with had his copyright claim denied. He subsequently made some money from a book about the project and selling the rights for a documentary.
That's a common legal hypothesis that's been suggested regarding AI. Has any US court accepted the 'paintbrush' argument as applying to AI-created works?
Yes, because it's faster and cheaper. You can now just outright create thousands of spam articles in an hour, or less, with only one person using prompts. SEO spam is already horrendous and is making search engines worse, publishers using this outright spam to both game SEO and advertisers is not great, imo. The problem is that there is plenty of backlash on bigger players doing this, but no one cares about smaller players doing it. Eventually that puts bigger players at a disadvantage, and they'll have to also start doing it enmasse. I think that's why we see them all slowly dipping their toes.
There is also a ton of stuff that has more or less formulaic sections where a person can cut and paste paragraphs that they could have written had they spent more time. This is what I've found ChatGPT/Bard somewhat useful for. I may reject. I will certainly selectively extract sections. I'll augment. But they can give me some useful material to provide background and otherwise flesh out the meat of a piece.
I’ve been considering this as the “complexity crisis”.
The 1960 high school dropout could work at Jiffy Lube and excel. Even to one point owning the store.
The 2020 high school dropout at Jiffy Lube has the attention span of a goldfish, pumped up on drugs his entire life, needs to know between this tool for this vehicle, that this brand needs this, that there are now 200 different oils, and no matter what happens even if he did excel that the store will be sold to a VC firm that will knock it down to put up a cheaply made 6-plex where three of the units are full time tax-free Airbnb rentals.
We have/are absolutely exceeding the ability of everyone, let alone the below average.
Half the time I look at what I need to know and wonder how most people are getting by.
> After we reached out with questions to the magazine's publisher, The Arena Group, all the AI-generated authors disappeared from Sports Illustrated's site without explanation. Our questions received no response.
Probably better to just admit to it to avoid the usual Streisand effect.
Every time an SI article finds me its happens to be by a confirmed human, however its tends to be the biggest fluff piece as if almost dictated by the persons agent.
Oh well, we were always a Sport Magazine household anyways, better writing.
I personally don't mind if they use AI to augment human writing/introduce interesting personas instead of the formulaic intro-bulleted list-conclusion articles from ChatGPT
Just a reminder that Ronald Reagan got his start making up baseball games: he'd read the ticker with the bare bones of what was up (Joe X at bat strike ball ball foul Joe X on 1st Bob Y at bat) and then dramatically pretend to be at the game for the benefit of the listeners: "Joe X strides to the plate, swinging his bat. He swings...strike! The pitcher, impassive, looks at the ball...and like lightning throws another pitch, but the ref calls it wide."
We need an aggregator that filters out such “news” agencies. Content written by ai? Garbage. Equally we need means to protect genuine content - perhaps text DRM.
https://www.espn.com/nba/preview/_/gameId/401584885
"The Associated Press created this story using technology provided by Data Skrive and data from Sportradar."
Example paragraph:
"The two teams match up for the second time this season. The Nuggets defeated the Clippers 111-108 in their last meeting on Nov. 15. Jokic led the Nuggets with 32 points, and Paul George led the Clippers with 35 points."
100% generated from the stats table, and totally boring and devoid of life. Horrible.