Login
You're viewing the functional.cafe public feed.
  • Nov 5, 2026, 11:11 PM

    One of the main objections I have to "large model" AI is that it is a form of piracy that fails to serve as an accurate form of archival. vis.social/@bleeptrack/1175600

    Every AI scraper develops an archive of copyrighted works that would put any library on earth to shame. Then they only let you access it it horrifically garbled, with semantic noise added, and through a straw

    💬 19🔄 226⭐ 412

Replies

  • 💬 1🔄 2⭐ 16
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 0
  • Nov 5, 2026, 11:52 PM

    @mcc

    Chokepoint Capitalism mind melded to Surveillance Capitalism and fossil fuel funded technofascism.

    💬 0🔄 1⭐ 7
  • Nov 6, 2026, 12:53 AM

    @mcc@mastodon.social It's frustrating to find stuff with too given how it works internally and dnesn't understand the material. I try and search up a schematic and the google AI I can't get rid of wants to direct me to useless places with made-up answers

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 5
  • Nov 6, 2026, 2:43 AM

    @mcc this clicked some things into place in my brain hard enough that I want to scream now

    💬 0🔄 0⭐ 6
  • Nov 6, 2026, 2:59 AM

    @mcc I'm not a big fan of using copyright law or intellectual property as an argument itself (***many obvious caveats here) but this is good. They're the antithesis of shadow libraries (archival, "working around" broken laws).

    Words are hard, and I perfectly understand why people use copyright infringement as an argument too… Ok I forgot what I wanted to write it's been a week

    *wrong one is prosecuted heavily for piracy

    💬 0🔄 0⭐ 1
  • Nov 6, 2026, 3:29 AM

    @mcc look at us, preserving these books. we're preserving them so hard that we're physically shredding them!

    what? no! we were FORCED to shred them, by this ASSHOLE judge who said it would be copyright infringement otherwise! blame HIM, not US, who could have simply not trained on these books and chose to shred them anyway!

    💬 0🔄 0⭐ 3
  • Nov 6, 2026, 4:02 AM

    @mcc what if a bt client that always trains a (toy) transformer model as it downloads anything, this ought to now have a lot of legal status now right?

    💬 1🔄 0⭐ 4
  • Nov 6, 2026, 11:16 AM

    @PsySal
    I know adding a toy transformer model to a BitTorrent client is not a serious suggestion, but let me pretend for a second it was! :blobfox3c:

    There’s two major differences that make this non-comparable, really:

    1. They obtained the books legally (or at least enough of them so they can plausibly pretend to for the rest, since there is no per-work papertrail), with BitTorrent you generally do not so you have no legal basis of processing to begin with
    2. They do not share the obtained works (sharing the GNN outputs are exempt since they are considered non-copyrightable in this framework), unless your client is purely a leeching client (like, scan and burn destroy of books is) you are engaging in illegal replication

    Sooooo, not really the same no.
    @mcc

    💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 2
  • Nov 6, 2026, 4:57 PM

    @PsySal @curiousicae The actual flaw in the joke plan is that running the bittorrent client may turn out to be the ONE THING that the AI scraper companies actually get in real trouble for.

    But not enough trouble. It will be a fine.

    This means that the real hack to being able to commit any crime you want is not LLM training, but money

    💬 2🔄 0⭐ 8
  • 💬 0🔄 0⭐ 0
  • Nov 6, 2026, 9:13 PM

    @mcc @PsySal @curiousicae more evidence that we need to change copyright laws to accommodate the existence of LLMs. They can only get in trouble for honestly sharing, not for plagiarizing. That's even more fucked up than them not being able to get in trouble at all

    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • Nov 6, 2026, 4:24 AM

    @Wiz4rd_of_Aus Theoretically anything you feed into the LLM goes into the context window, not the training set. Unless the AI companies are lying to us. Would AI companies do that? Just lie?

    💬 2🔄 0⭐ 6
  • 💬 0🔄 0⭐ 3
  • Nov 6, 2026, 11:02 AM

    @mcc
    At least OAI apparently has a premium subscription who’s only distinction from the base one is that the premium one is not used for training, so it’s not even a lie there.
    @Wiz4rd_of_Aus

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 6, 2026, 5:37 AM

    @mcc Also, when they scrape the real world, i.e. books, magazines etc. they FUCKING DESTROY THEM IN THE PROCESS, which I'm pretty sure is the opposite of preservation...

    💬 0🔄 0⭐ 1
  • Nov 6, 2026, 6:19 AM

    @mcc first they steal it, then they go bankrupt, then they sell the data for pennies

    💬 0🔄 0⭐ 3
  • Nov 6, 2026, 6:38 AM

    @mcc Nailed it. We need laws against this.

    The (imo) clear sign that this is intentional is that LLM interfaces are offered by those powers that disable access to the non-garbled indexes.

    This begs the question of the goal of making access to public information harder.

    The only answers I can see are ethically appalling.

    Regardless of the question of the use of llms, the greater good of disabling free direct search is obviously non-existent.

    💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • Nov 6, 2026, 9:37 AM

    @mcc it's just wild to even think about the scale of data hoarded by these greedy companies only to gatekeep it behind a restrictive API or chat interface that actively takes away the context of original works by talented people.

    💬 0🔄 1⭐ 2
  • Nov 6, 2026, 11:22 AM

    @mcc
    I’m kinda hoping there will be a “leak” of one of those company’s archives (the source data, not the models ofc) once they go bankrupt. They couldn’t have deleted any of it since it’s needed for training newer models, so maaaaybe pretty please? Some disgrunted (ex-) employee maybe? You just forgot the password on that fileserver and “suddenly” some archivists noticed (totally not tipped off by some untracable email)? :blobfoxlurkaww:
    @cinebox

    💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • Nov 6, 2026, 6:25 PM

    @mcc It's a vexing problem with modifications of "dead" computer games too. People wrap the game with AI optimizations, then add in modifications over the original work, and basically combine both into a new game based on prior work ostensibly without permission of all who created it.

    (example: openomsi.org/)

    💬 1🔄 0⭐ 0
  • Nov 6, 2026, 6:27 PM

    @allanb in the long run i think this will be bad for preservation. with a romhack, you changed the original, but you know what you did. "AI" recompile artifacts constitute randomly generated false versions of vintage games where the developer does not and does not have the ability to actually know what exactly they changed, and to the extent they do know is potentially not honest about what or how much they changed.

    💬 1🔄 0⭐ 1
  • Nov 6, 2026, 6:34 PM

    @mcc This particular game has a huge amount of 3rd party community maps that were painstakingly created by hundred(s)? of people over the years. The maps can all be re-used in the AI version, but are almost unrecognizable due to "aesthetic" changes added-in by AI. IMO optimizing a vintage game to modern computers is one thing -- changing it's appearance and play is another thing entirely, defacing original art. I doubt the AI devs can separate the two products.

    💬 1🔄 0⭐ 2
  • 💬 0🔄 0⭐ 1