Kindle novelists are using ChatGPT(theverge.com)
theverge.com
Kindle novelists are using ChatGPT
https://www.theverge.com/23520625/chatgpt-openai-amazon-kindle-novel
5 comments
Given the amount of absolute garbage I’ve come across in Kindle unlimited, this doesn’t surprise me. I’m not denigrating the service or writing in general, I’m just aware of how much it is becoming a dumping ground due to lack of curation. I’ve even seen advertisements on websites promoting how you can make easy passive income by outsourcing a book and selling it on kindle unlimited … ChatGPT is just removing the cost to outsourcing which was an issue until now due to the cost of using GPT3, because honestly some of the stuff I’ve found on kindle unlimited is unquestionably written by people with a thin grasp of English and to suggest they were even read before publishing let alone actually edited would be an insult to anyone that has ever done the job of editor. GPT3 while prone to waffling on like grandpa Simpson talking about an onion on his belt… it at least waffles on using what I would call coherent readable English.
Relevant high quality and interesting video that dives in to the world of get rich quick book dumping schemes:
https://youtu.be/biYciU1uiUw
What I worry about is that GPT-written books are going to be well enough written on the surface to pass the "read ten pages before deciding if I buy" test. These are the going to be the books where you get 70 pages in before realizing there isn't a coherent story there.
If the book is poorly written and it's obvious, then it's probably not going to gather much of a following but it's not going to hurt anyone. The next two years of book spam are going to be grammatically impeccable and require so little effort that a few people are going to make large amounts of money, at the expense of legitimate self-publishers who'll have an even harder time getting noticed at all, selling fake books on Amazon until the hammer comes down.
After that, traditional publishing becomes the target. By 2025, Amazon will have the fake book problem largely solved--it'll still be possible to get fake books through, but at some point it'll be harder than to just write a real one--while New York publishers are running the third-party software the MBAs bought because it was cheap. Meanwhile, even though the AI-written books won't be good in the sense of James Joyce, they'll be good enough to best-sell. And, of course, there will be prestige associated with being the first one to pull it off. If you make a living selling fake books on Amazon, you don't admit to it. If you get the New York Times to review your book, and if you get a few celebrities to include it in their book clubs, and then reveal (with proof) that it was written by AI, you become a hero among the 90 percent of the US who hate the New York Times (and, to a lesser extent, traditional publishing) because their kids will never get hired by places like that. Selling fake books on Amazon is shameful; selling a fake book to a publisher for a seven-figure advance, and all the big ticket reviews and celebrity endorsements included in the package, is hilarious. Of course, 99.99% of attempts at this are going to fail somewhere in the process--the hardest part is going to be creating fake identities [1] so you can query the same agents hundreds of times--but there are going to be way more than 10000 tries.
It actually wouldn't surprise me if someone writes a real book, gets it into traditional publishing, makes the bestseller list, and then tries to go back and say that it was written by AI, just to become a national hero (as opposed to just another guy who was an NYT bestseller for a few weeks). Which is why I hope the first person to actually pull it off does a good enough job of documenting his process that he can prove that it really was a GPT novel... so he can't be accused of this.
----
[1] Gray hat uses of LLMs, such as creating fake personal identities for social proof and credibility purposes, may be more widespread than the black hat purposes we're all scared of. It's now really easy (seconds vs. hours) to get the 3- or 4-digit karma count necessary to not look like a sock puppet. In the past, the gray hats brought Reddit and HN accounts; in the late 2020s, you'll be able to buy whole "people", with deepfaked audio, video, and text presence all over social media, for $100 a pop.
If the book is poorly written and it's obvious, then it's probably not going to gather much of a following but it's not going to hurt anyone. The next two years of book spam are going to be grammatically impeccable and require so little effort that a few people are going to make large amounts of money, at the expense of legitimate self-publishers who'll have an even harder time getting noticed at all, selling fake books on Amazon until the hammer comes down.
After that, traditional publishing becomes the target. By 2025, Amazon will have the fake book problem largely solved--it'll still be possible to get fake books through, but at some point it'll be harder than to just write a real one--while New York publishers are running the third-party software the MBAs bought because it was cheap. Meanwhile, even though the AI-written books won't be good in the sense of James Joyce, they'll be good enough to best-sell. And, of course, there will be prestige associated with being the first one to pull it off. If you make a living selling fake books on Amazon, you don't admit to it. If you get the New York Times to review your book, and if you get a few celebrities to include it in their book clubs, and then reveal (with proof) that it was written by AI, you become a hero among the 90 percent of the US who hate the New York Times (and, to a lesser extent, traditional publishing) because their kids will never get hired by places like that. Selling fake books on Amazon is shameful; selling a fake book to a publisher for a seven-figure advance, and all the big ticket reviews and celebrity endorsements included in the package, is hilarious. Of course, 99.99% of attempts at this are going to fail somewhere in the process--the hardest part is going to be creating fake identities [1] so you can query the same agents hundreds of times--but there are going to be way more than 10000 tries.
It actually wouldn't surprise me if someone writes a real book, gets it into traditional publishing, makes the bestseller list, and then tries to go back and say that it was written by AI, just to become a national hero (as opposed to just another guy who was an NYT bestseller for a few weeks). Which is why I hope the first person to actually pull it off does a good enough job of documenting his process that he can prove that it really was a GPT novel... so he can't be accused of this.
----
[1] Gray hat uses of LLMs, such as creating fake personal identities for social proof and credibility purposes, may be more widespread than the black hat purposes we're all scared of. It's now really easy (seconds vs. hours) to get the 3- or 4-digit karma count necessary to not look like a sock puppet. In the past, the gray hats brought Reddit and HN accounts; in the late 2020s, you'll be able to buy whole "people", with deepfaked audio, video, and text presence all over social media, for $100 a pop.
Given how little Amazon have done to protect the customers of physical goods from getting garbage, I don’t think they would even bother looking for the metaphorical hammer let alone dropping it on garbage book writers filling the store with what is basically spam today and GPT nonsense tomorrow…
Case in point: You can find numerous instances of people (multiple people often do this for the same book leading to multiple entries) uploading public domain works of the US government such as NASA NTRS pdfs, or military history compilations that were done by government military historians and are thus also public domain… all uploaded to Amazon by someone hoping that their Amazon link shows up higher in the search results (or gets the top spot as a suggested shopping link) than the original free public domain PDF. Amazon seems ok with this.
Another case: Amazon happily allows multiple garbage grade copies of public domain books to coexist in the kindle store (last time I looked) there were plenty of old project Gutenberg text dumps run though some kind of txt2mobi pipeline and just left on the store as a dollar, or even if they bothered to try and have a nice cover art/jacket they usually set the price as a few dollars… all for a public domain book that should be free (and is since there’s also usually someone who has uploaded the same book and listed it for free)
I think the mantra at Amazon is “more books = bigger kindle store = more value for customers using the kindle ecosystem = more good” with zero consideration of how they could be slowly undermining the consumer trust in their platform. This feels to me like the same management playbook at work in the physical goods space where Amazon are facing a rising tide of counterfeit goods and they seem structurally incapable of solving these problems… if management cared something would be different. Why will this be different for ebooks?
Case in point: You can find numerous instances of people (multiple people often do this for the same book leading to multiple entries) uploading public domain works of the US government such as NASA NTRS pdfs, or military history compilations that were done by government military historians and are thus also public domain… all uploaded to Amazon by someone hoping that their Amazon link shows up higher in the search results (or gets the top spot as a suggested shopping link) than the original free public domain PDF. Amazon seems ok with this.
Another case: Amazon happily allows multiple garbage grade copies of public domain books to coexist in the kindle store (last time I looked) there were plenty of old project Gutenberg text dumps run though some kind of txt2mobi pipeline and just left on the store as a dollar, or even if they bothered to try and have a nice cover art/jacket they usually set the price as a few dollars… all for a public domain book that should be free (and is since there’s also usually someone who has uploaded the same book and listed it for free)
I think the mantra at Amazon is “more books = bigger kindle store = more value for customers using the kindle ecosystem = more good” with zero consideration of how they could be slowly undermining the consumer trust in their platform. This feels to me like the same management playbook at work in the physical goods space where Amazon are facing a rising tide of counterfeit goods and they seem structurally incapable of solving these problems… if management cared something would be different. Why will this be different for ebooks?
There are some companies now focused on this kind of assisted fiction like https://verb.ai/