Scammy AI-Generated Books Are Flooding Amazon
wired.com
wired.com
Instead look for examples of real usage of the word in use e.g. https://youglish.com/
English is my mother tongue and I often need to look up how to say and use phrases from other English speaking regions e.g. "nowt" recently.
And take care to pick your accent - particularly American or British. "can't" is particularly hard to say properly in my English and US people misunderstand the way we say it.
What'd you call me?
Or people with thick foreign accents, which is just baffling.
Really damages the perceived quality of the service when it can be gamed like that.
That's a really common genre of video on Youtube. If you Google "how to pronounce 'Stranger Things'", you'll find dozens, including this channel with separate videos for "How to pronounce Stranger Things" (https://www.youtube.com/watch?v=ZygvdKNT15o) AND "How to pronounce Stranger Things 3" (https://www.youtube.com/watch?v=mD4Zae1rOgY).
A ton of people jumping on the bandwagon promising insane profits from AI. But if you thought for more than 1 second, if they themselves could make this much profit why would they be trying to convince other people to do it.
The reality is they are likely the ones making any money by building up a following making these promises.
Often times this stuff is scams that stopped working, so the scammers pivot to selling the non-working scam to others.
There was a YouTube video a couple years ago that exposed the scam of a couple of twins that were selling SEO'd bottom-of-the-barrel ghostwritten books on Amazon. Instead of using generative AI, they used the cheapest possible online ghostwriters. Apparently Amazon got wise and banned them, so they pivoted to selling their "system."
Edit: here's the video: https://www.youtube.com/watch?v=biYciU1uiUw "Contrepreneurs: The Mikkelsen Twins"
The typical scheme that gets people rich since "Rich dad, poor dad", but now it's exponentially easier to do. It's crazy
The hustle? Spam amazon affiliate links, lol. Of course the people pushing this never bother with pushing affiliate links, just trying to sell you the guide on how to do it.
It was a bit before the AI boom, back when the bedrock of the scam economy was mostly gig platforms.
like, it requires a lot, but not exponentially
I had previously thought that the guy was somewhat smart and had it together. After that, I never looked at anything he did the same way again. The nice clothes and car? Yeah, no way he wasn't doing all that on a load of debt.
That said, all his up-to-date management-speak and being 'with-it' when it came to the latest corporate craze, well, I hate to admit it, but it made him a better boss. He was at least trying really hard. Were a lot of the methods just hot air? Yeah, of course. But it was a lot better than a boss that actually thought that they knew what they were doing. My boss just pretended but was otherwise terrified of being left out, and at least he was hungry that way.
Still, after he boasted about his dropshipping business, I knew the guy was mostly hot air and his ventures would change with the wind. I'm glad that particular get rich quick scheme is (mostly) over.
Even pre-AI lots of childrens book at the lower age range/illustration category are basically low effort junk, this is generally true for most/all books for new children. They use illustrations bought wholesales and copy and paste text into them.
This conversation over Christmas dinner set off so many red flags, that I could only bite my tongue and nod.
I sure hope this blows over quickly.
They made a token effort to curtail some AI spam by limiting how many self-published books you can upload per day to 3, but that's still laughably high.
Real question is whether they'll start making their own 'AmazonBasics' branded AI-generated dogshit, just like they do to undercut/displace successful 3rd party products.
color me surprised.
Fast, free, or fact: pick any two. The current internet is fast and free, but is increasingly non-factual. An alternative emphasizing meaningful facts and observations will almost necessarily be slower.
We might even be able to use pressed bleached dead trees as a backup medium.
I only have so much time in the day. I can only consume so much content. I don't need AI or machine learning to filter a firehose of content.
My friends provide a human scale stream of recommendation that I trust much more than a profit driven algorithm.
Also, stop using Amazon for delivery of said content as they are part of the issue.
Those names, brands that have been established for decades, are how I curate. I feel reasonably certain BBC and FX will continue serving valuable content for the foreseeable future. Random YouTube channel I've never heard of or all-caps company name on Amazon not so much.
I hope so but there are forces in the UK (and I suppose outside it) who see the BBC as "unfair" competition and will continually attempt to defund/destroy it.
I haven't watched terrestrial tv in a decade or more, pretty sure there's a generation of UK kids who will never bother because youtube/twitch/netflix is better as far as they're concerned.
That said this won't happen and platforms will shift from metrics like views to things like number of uploads or "shares" to show "growth".
And how fast can I throw content at that AI-checker and get an AI vs human score?
It shouldn't matter if something was written by AI or human if it is high quality , fact-checked. But I think AI (or the way we use it) still needs some improvement for that.
"still" ? This isn't even in the purview of current technology, afaik.
Why not 3 books a month? Or even a few a year? Even extremely prolific authors rarely write more than a few books a year. This is a ludicrous rate-limit that seems designed to not actually do much (so Amazon can still get their commission) while giving the appearance that it does. It's hard to take this in good faith.
Covid only accelerated this trend as people began to look for other ways to generate income online. Just look at the Gamestop and AMC amateur stock trading craze or the explosion of flipping sneakers or Pokemon cards for income.
A huge percentage of the internet consists solely of people looking to make a quick buck and/or establish some channel of passive income. The internet itself has been redefined as this weird vehicle for capitalism because we're careening towards a weird transitionary period in history as the rise of AI takes hold.
The only thing that can solve this is UBI or some kind of shared equity system where huge megacorps primarily operated by AI generate trillions and trillions of dollars in wealth that needs to be distributed to the populace somehow.
Godspeed as we enter this new age.
HN still feels like the old internet for the most part, at least. :)
I believe it wasn't only journalistic gatekeeping that prevented this for so long, but simply a lack of affordance for willing parties to do those things. High-speed Internet has made everything so much easier.
My feelings about the rise of streaming video are also conflicted. It has given us high-quality longform content on topics that hadn't been covered before, but all those other people's passions and hobbies take up way too much time if you as a consumer become deeply invested in them (not even counting the conspiracy genre and its implications). It's easy to do partially because YouTube facilitates that, but I think in the end people want to consume all this information in the end, for some definition of "want". And dozens of 6-hour streams of the latest game or another will keep being uploaded every day.
And there's this other thing that I keep coming back to. Someone I know was really into this engineer creating fake Apple product boxes that recorded thieves stealing them and being glitterbombed. They found it hilarious, but I was sort of confounded. It was the conversion of a few people's bad choices into entertainment and ad revenue, coated over with the intellectual sheen of detailed makerspace engineering diagrams and wit mixed with a moral high ground. Though what the thieves did was not right, there was something smug and passive aggressive about the entire thing that rubbed me the wrong way. And this was presented as just a humble guy with engineering experience creating his content and getting people interested in STEM. Anyone has the capability to do this sort of thing now, not just a media machine with questionable motives. It's just regular, driven people with questionable motives now.
What's been bothering is the layers of skimmers that are dominating all those mediums - the top youtube channels now are repeaters, the top tiktok channels are fake accounts copying the content, there's layers of "reactors" (they call themselves that) that add absolutely no value, but polute the whole timeline, and so on. This is mostly a newer phenomenon - as if the internet is quickly turning into a multimedia version of the mail system - it'll snuff out actual creativity/usefulness and slowly decay into spam spam spam spam.
Current discoverability outside people-you-actually know seems to be in a nosedive, so it's hard to see this getting better any time soon.
If you look for coloring books on Amazon it's basically all AI generated.
Maybe it would help here too?
What other alternatives could mitigate the problem?
Pretty sure it’s just a bunch of journal articles patched together, each one as a chapter.
(Amazon refused to let me publish my books.)
Will we need to come up with a way to authenticate non-AI content?
Publish it prior to 2022.
Add that to the fact that these systems are built by scraping work without permission and these turn into headwinds that push people away from creative pursuit. Rather than bringing new people into these communities, the people who actually know what they're doing just get trampled by the uninformed.
None of these dynamics really exist with photoshop. While photoshop has tools that can speed up the artistic process, you need a base understanding of how to illustrate to actually use it effectively. It's truly a tool, as opposed to a slot machine you just give a prompt to.
As for the "scraping" issue -- that is a boring sideshow that will be decided by lawyers -- is it fair use or not? Your or my opinion on that whether it "should be" is irrelevant. But even if the courts decide that training generative AI on data the trainer doesn't have permission to is illegal, this wouldn't "stop generative AI" as many seem to think. For example, Adobe is training its models only on public domain works and works it has the rights to.
No it isn't, because many folks live in democracies, and this is a decision that should actually be made and implemented into new law, instead of leaving it to courts to try to apply some outdated laws to a situation they could not possibly have been written for.
Oh wait, I did call it "AI-generated crap", but I am open to the idea that not all AI content is crap.
I mean, I get that you can't control what the entire population does but, AI is a powerful tool that enables a lot of incredible things we may have never imagined being possible before. And it's not even out of its infancy yet. But people are ruining it.
Similarly, a decentralized currency or decentralized ledger is a pretty great idea for a lot of things. The execution could use work but there are a lot of benefits to that. But it ended up being a way for a bunch of bros to scam a quick buck and now it's a joke.
I'm not here as some jaded blind fanboy or crypto bro. I legit just hate how people ruin everything for the rest of us. I want to be able to enjoy cool new tech and interesting new ways of achieving things or learning. I want to see innovation. But what we get is "everything is trash because people are trash and will do anything to get money"
If you're generating AI content that's better than any book you've ever read, in the _current_ state of what's available, I'd suggest reading some different books.
I believe we'll find a balance, as with anything. There will always be people publishing junk, including AI-generated junk. But many authors will learn to use AI to supplement their work and produce a high-quality product.
With all due respect, you're probably kidding yourself about the quality of your AI generated book.
4 is somewhat more interesting, and I've seen worse from actual humans, but… roll a d6, and on a 1-5 it's terrible, roll a 6 and you get something that's 90% fantastic and 10% just a bit off. For example, this story: https://github.com/BenWheatley/Fiction/blob/main/The%20Silen...
I like it, but if I really examine the thing, the details don't quite line up right. The humans were supposedly primitive, yet "[t]heir cruisers absorbed salvos that would've shattered lesser vessels"?
And sure, there are a million oddities like this in "proper" fiction[0]; but this… still feels like Stable Diffusion in text form. It's cool, it's a technological marvel, it's still, ineffably, not quite right.
[0] Even ignoring that the top speed in the pilot episode should've gotten them home in 7 years rather than 70, Voyager had a bajillion ways to get home sooner — for example, putting the crew in stasis so they didn't age, going at a normal speed, and then using any of the time machines Starfleet keeps tripping over.
To me, it's fine that it has no named characters, no dialog. One of the most grating things in plots with aliens is the names, no matter if they're American names like the spiders in A Deepness in the Sky, or the Star Trek pattern giving us random apostrophes and internal capitalisation, where "Q'ting'pAh" could easily be either Klingon or Vulcan.
[1] Not all of the output of 4, like I said it's mostly bad, but that I like this particular result.
This is a bit of a tangent, but: you missed an important point in that book with some broad implications. Those weren't the spiders' names, those were the translators' deliberately humanized names for the spiders.
(And, to bring the discussion back home: no language model would have ever come up with a clever narrative trick like this.)
Just because it had a good in-universe reason didn't stop it being grating.
How much editing are you adding/subtracting to the content?
How granular are your prompts (e.g., prompting for roughly each new paragraph, ~page, chapter, or larger)?
How much are you finding you need to repeat/retry to get good output for your project?
I can't share my exact prompt style since I believe I've found a reliable way to generate these books en masse with minimal factual errors.
If there are factually incorrect things to the books hallucinated by LLMs, they're not non-fiction.
>...minimal factual errors
This is still really bad when pretending to be a book portraying historical events...
Are there any AI generated books out there that have found an audience and built a fan base?
I don't doubt that it's going to happen but I'm skeptical if we are all that close.
There are some very successful authors out there that many would say are awful, so that's why I asked for books with a fan base rather than good books.
I'd be surprised if the book can follow a consistent plot rather than hallucinating characters halfway through.