• 0 Posts
  • 10 Comments
Joined 2 years ago
cake
Cake day: September 6th, 2024

help-circle



  • I think what they’re saying is that the AI companies don’t need to rinse and repeat. They’ve already created solid programs for how to make LLMS speak English and demonstrate basic reasoning. You don’t need to retrain that part every time. Once you’ve trained “English.exe,” you can just copy it endlessly.

    Maybe with enough time, the hard-coded English of the LLMs could become increasingly anachronistic and sound old-fashioned and formal to most ears. But for that kind of for slow maintenance you could just pay people to write examples of modern language and train it on that.


  • I envision the opposite:

    In the future, I imagine authors will go to great lengths to prove their works to be authentically human. It may become standard practice for authors to live-stream their writing. Literally a years-long stream of them sitting at their desk. Comments turned off.

    Perhaps libraries could even offer this as a service. Offer a space where authors can come write in public and have it simultaneously live-streamed. Maybe it becomes the norm for authors to mostly write publicly on a platform like Twitch, but to occasionally write in public spaces where they can be physically observed.

    This will not be a legal requirement to publish a book. But no respectable author would do otherwise.



  • Why would they scan that stuff though? That kind of mass market (human made) slop is easily available in digital form. We know the AI companies have engaged in mass digital piracy, including running massive torrenting operations. So if a digital copy exists, they probably already have it. And even if they have to buy it, purchasing an ebook is a lot cheaper than buying a physical one, shipping it, paying someone to scan it, etc.

    I would think the old, the out-of-print, the rare, and never-before digitized are the only things worth buying and physically scanning in 2026. Everything else has already been scanned or was born digital-native.


  • Are you sure they’re digitizing Book By the Yard quality material?

    Consider this. Purchasing, shipping, and physically scanning a book is the single most expensive way for an AI company to acquire the text of a work. It’s been widely documented in court cases that these companies engage in mass IP theft. They’re literally running massive torrenting farms, grabbing copies of every film, song, book, etc. that they can get their hands on.

    What kinds of texts are most likely to be found pirated on the internet? It’s the mass market stuff. The common stuff was digitized long ago. They can just download that. They can pirate it. They can buy the ebook. There’s no need for OpenAI to purchase and destructively scan the works of Steven King. No shade on the man, but his works aren’t exactly hard to find. I’m sure they can just find a torrent.

    There’s little value in scanning the mass market books that are produced in enormous quantities. They probably don’t buy “Book by the Yard” books, because they already have digital copies of those.

    But the rare, long out-of-print stuff? The stuff that you actually cannot find a legal or illegal copy of online? That’s only stuff worth paying money to buy, ship, and scan.

    Doing anything physical, especially at scale, is slow and expensive. I would guess a good portion of these books, perhaps an outright majority, have simply never been digitized.