"I’m sure the exposure translates into real sales:" - and indeed, I came across the Twitter / X account of one of the authors of the "100,000 whys" slop-books. It's not the category bestseller shown above, but they still seem to be doing well:
Sure, in context. No one was trying to tell them that it's real life.
But I'm not saying it's the end of the world; there are far worse things kids and adults can be exposed to. At the same time, this contagion will spread, and it's also entirely self-inflicted and pointless. It's not like we didn't have the exact same books before, except done more competently. All the Dorling Kindersley encyclopedias, etc.
Take scientific research: published papers are assumed correct provided the logic holds up and no ones shown otherwise. This system works because since most people write in good faith, fabricated data is rare enough to be caught manually.
LLMs make it extremely easy to crank out slop, at a rate no human could even hope to verify (which requires repeating the experiment) and therefore, pose an existential threat to the system.
A similar thing applies to books: because writing a book is hard, most information found in published books can be trusted until proven otherwise. Of course, bad books exist, but there aren't very many and they are easy to weed out... until now.
I'm, like you, of two minds... On the one hand I really DO believe that, in a fair and free market, cream rises to the top and crap sinks to the bottom.
On the other, are "fair" and "free" anything but aspirations in this fallen world? Seems like... and if that's the case, how often does intervention make it worse rather than better? I'm not an anarchocapitalist by any means, don't get me wrong, but market incentives don't reward intervention.
At the end of the day, I just don't want a generation of grandchildren to think that some kids have trunks dangling out of their massive Neanderthal chins. And I hate this kind of lowest-common-denominator, crank-out-the-crap garbage. But I don't hate the system that incentivizes it.
The way I see it, the problem isn't that the models aren't ready yet. The problem is that people are publishing AI written books without even looking at them. If they'd taken the time to do a simple human review (like you provide here) these problems would have been avoided
Except that much of the improvement in the past few years hasn't been to the factual accuracy of the invented reality, but it's believably. In the past few years, they've gone from stringing words together to making something that almost looks like a logical argument.
Granted, modern chatbots just google any question you give them and plagiarize the results. This usually produces a correct answer, but it's contingent of the quality of those results... something which is actively getting worse with the proliferation of AI-slop.
I don't think I'd want a book composed from from the first google result to common questions, which in recent days seem to be entirely slop marketing blogs and equally low effort tiktok videos.
(Of course, this is all speculation and I have no way to prove what will happen in the future. But recent history seems to support that better LLMs -> More AI slop)
It was always the fundamental problem of Amazon - books can be published that nobody ever read before. You can publish a book describing location of buried pot of gold, and if you have no sales your treasure is safe!
I mean, yes, but also no? "The problem isn't our patented Bathtub Toaster. The problem are people who take our patented Bathtub Toaster into the bathtub."
There are financial incentives to pollute the world with cheaply-produced slop. Before LLMs, this required more effort and was easier to filter out. LLMs make it possible for almost anyone to produce spammy content on a galactic scale. The content also has higher production qualities, so it takes far more effort to figure out what's what.
I do agree that AI has value and some areas will absolutely benefit from the use of AI, but so far, none of the examples I've seen, especially these, would have convinced me.
I want a copy of your book, as you are a known entity with a... Soul? Morals? A compass? AI, unlike people, can never be held to account. Whether it's a self driving car running someone over, or a mutant ear/nose diagram, there's no way to ever improve anything.
"I’m sure the exposure translates into real sales:" - and indeed, I came across the Twitter / X account of one of the authors of the "100,000 whys" slop-books. It's not the category bestseller shown above, but they still seem to be doing well:
https://lcamtuf.coredump.cx/blog/slop-author1.jpg
https://lcamtuf.coredump.cx/blog/slop-author2.jpg
Further funnies: in one of the books, someone accidentally asked the model to generate vector images instead of bitmaps, which ended poorly: https://lcamtuf.coredump.cx/blog/whys-vector.jpg
The heck… the dystopian future arrived much earlier than I thought.
It should be fine. Kids are resilient. I showed both of my kids HR Giger’s art when they were 5 and they seem fine.
Sure, in context. No one was trying to tell them that it's real life.
But I'm not saying it's the end of the world; there are far worse things kids and adults can be exposed to. At the same time, this contagion will spread, and it's also entirely self-inflicted and pointless. It's not like we didn't have the exact same books before, except done more competently. All the Dorling Kindersley encyclopedias, etc.
No I was joking! Those alien images still give me nightmares.
I hate the sound of these words but: it should be regulated/penalised by law
Take scientific research: published papers are assumed correct provided the logic holds up and no ones shown otherwise. This system works because since most people write in good faith, fabricated data is rare enough to be caught manually.
LLMs make it extremely easy to crank out slop, at a rate no human could even hope to verify (which requires repeating the experiment) and therefore, pose an existential threat to the system.
A similar thing applies to books: because writing a book is hard, most information found in published books can be trusted until proven otherwise. Of course, bad books exist, but there aren't very many and they are easy to weed out... until now.
I'm, like you, of two minds... On the one hand I really DO believe that, in a fair and free market, cream rises to the top and crap sinks to the bottom.
On the other, are "fair" and "free" anything but aspirations in this fallen world? Seems like... and if that's the case, how often does intervention make it worse rather than better? I'm not an anarchocapitalist by any means, don't get me wrong, but market incentives don't reward intervention.
At the end of the day, I just don't want a generation of grandchildren to think that some kids have trunks dangling out of their massive Neanderthal chins. And I hate this kind of lowest-common-denominator, crank-out-the-crap garbage. But I don't hate the system that incentivizes it.
Does that make ME crazy? Or is the world crazy?
The way I see it, the problem isn't that the models aren't ready yet. The problem is that people are publishing AI written books without even looking at them. If they'd taken the time to do a simple human review (like you provide here) these problems would have been avoided
Except that much of the improvement in the past few years hasn't been to the factual accuracy of the invented reality, but it's believably. In the past few years, they've gone from stringing words together to making something that almost looks like a logical argument.
Granted, modern chatbots just google any question you give them and plagiarize the results. This usually produces a correct answer, but it's contingent of the quality of those results... something which is actively getting worse with the proliferation of AI-slop.
I don't think I'd want a book composed from from the first google result to common questions, which in recent days seem to be entirely slop marketing blogs and equally low effort tiktok videos.
(Of course, this is all speculation and I have no way to prove what will happen in the future. But recent history seems to support that better LLMs -> More AI slop)
It was always the fundamental problem of Amazon - books can be published that nobody ever read before. You can publish a book describing location of buried pot of gold, and if you have no sales your treasure is safe!
I mean, yes, but also no? "The problem isn't our patented Bathtub Toaster. The problem are people who take our patented Bathtub Toaster into the bathtub."
There are financial incentives to pollute the world with cheaply-produced slop. Before LLMs, this required more effort and was easier to filter out. LLMs make it possible for almost anyone to produce spammy content on a galactic scale. The content also has higher production qualities, so it takes far more effort to figure out what's what.
Gahhhh!
Burn it with fire!
Damn.
I do agree that AI has value and some areas will absolutely benefit from the use of AI, but so far, none of the examples I've seen, especially these, would have convinced me.
I want a copy of your book, as you are a known entity with a... Soul? Morals? A compass? AI, unlike people, can never be held to account. Whether it's a self driving car running someone over, or a mutant ear/nose diagram, there's no way to ever improve anything.
Messing up kids, or teaching them to distrust unreliable authorities?
Never trust a three-legged cat