Contrepreneurs: The Mikkelsen Twins

> This is what I call meta advice. That’s what I call a grift. Your description reminded me of Dan Olson’s fantastic video on that type of strategy—where you make money by selling the idea that anyone can make passive income by doing something the seller doesn’t themselves do (anymore)—using the Mikkelsen Twins as the example. In Dan’s words (emphasis mine), the Mikkelsen Twins aren’t special, grifters like them are a dime a dozen. He chose them as the example “because they are of a type. They are a representative sample of a category of grift. And also because they’re kind of incompetent and that makes them entertaining ”. He’s not wrong. [this video]

Zoe Bee did a video essay[0] on her experience as a ghostwriter producing essentially the kind of junk that is now being automated with LLMs. In her case she was (somewhat unknowingly) paid to write unlicensed Minecraft stories. I think the video is worth a watch if you're not aware of the state of ebooks (especially on Amazon) prior to LLMs. It's also worth mentioning that this problem isn't limited to ebooks. There's also a cottage industry of mass produced minimal-effort audiobooks on Audible as part of various "passive income" scams. Dan Olson made a pretty good video essay[1] about one such scam where he actually played along for most of it and also gave it a try as a ghostwriter. [0]: [link] [1]: [this video] EDIT: Considering the most cited examples are Amazon/Kindle and Audible, I think parallels can also be drawn to the proliferation of no-name brand Chinese whitelabel dropshipping products on regular Amazon. Everyone already knows not to trust Amazon reviews but at this point it's hard to find reputable brands for products you're not already familiar with (e.g. which of the one hundred brands featuring near identical products are actually brands you might find in a retail store rather than a random name slapped on the product in the same Chinese factory?). LLMs will definitely make reviews even more untrustworthy but they might also help generating even more plausible copycat product descriptions and designs.

One thing to note is that if you're not on Amazon, your book will still be on Amazon. Cory Doctorow points out[0] that their copyright detection systems only detect reuploads of on-platform books, and people scamming Audible with duplicates of books are rampant. I suspect this is part of why there isn't a huge contingent of angry authors jumping ship, because they can't reliably deny Amazon their work anyway. You're supposed to be able to just tell platforms "no, take this down". That comes from DMCA 512, that thing that indie creators absolutely hate because its far more useful as a censorship tool than for copyright enforcement[1]. If someone else reuploads your book to Amazon you have to actually find that reupload, buy it to validate that it is your work, take it down (and dox yourself in the process), and then repeat, forever. Outright infringement isn't the only scam either - ghostwriting and reselling lightly-edited GPT-3 copy[2] is also rampant on the platform, and there's even people selling courses on how to resell fake books[3]. This may eventually destroy Amazon's brand reputation, but it takes a long time to fall down the Trust Thermocline[4] when you're a giant like Amazon. The $10 royalty punishment makes absolutely no sense to me and I do not understand what the hell Amazon's motive is there. Except perhaps to bluff-charge indie writers into devaluing their work ahead of time. It reminds me of how Apple pushed for $0.99 songs for so long that the RIAA was will

So far AI art is doing the opposite: increasing the amount of crap that people produce. The kinds of people who were once limited to underpaying ghostwriters to make spam books for Audible[0] can now chuck together a few prompts to ChatGPT and Stable Diffusion and get the same result. Yes, you have to be good at art in order to make good art with AI, but that doesn't matter when your goal is to create spam. The idea that artists should "do stuff for the joy of creating" is just plain insulting, too. They already do that . While artists would love to see, say, AI art models that were trained with licensed or public-domain data; training data theft isn't even their biggest concern. Their biggest concern is having the fun sucked out of their job as the artful minutae of drawing or writing is replaced with finding the correct combination of words to make Stable Diffusion draw the character you want with exactly the same details every time. It would be like if you worked at a PC building shop and one day the boss said "Actually we're just going to be an Apple authorized reseller now." The thing that's destroying artists' jobs being trained on their own work is just insult to injury. To be clear, though, AI doesn't "steal and resell content" in the vast majority of cases, either. Regurgitation is a thing, but the cause is duplicate data in the training set making it advantageous to memorize a few images to improve loss metrics. Most diffusion model architectures are not big enough