Login
You're viewing the front-end.social public feed.
  • Oct 8, 2026, 6:28 PM

    I've warned before that any FOSS project incorporating AI output is doing so at their peril.

    Previously people leaned on a federal circuit ruling that was *declined to be taken up by the supreme court* interpreting it as a free-for-all to LLM license-wash any project in translation, or really to incorporate any such materials and they would be public domain.

    But that doesn't mean we have an answer! We've yet to see!

    Annnnnnnnnd: I think we're likely to have that perspective upended soon.

    The third circuit issued a recent ruling in the *opposite direction* of the previously well cited federal ruling, that AI training on copyrighted material is NOT fair use: courthousenews.com/ai-training

    Annnnndddddd.... with a bunch of people trying to make LLM-generated decompilation replacements for Adobe's entire suite, we're likely to suddenly see corporations rear up angrily.

    Just wait till someone tries to vibewash a Nintendo game decompilation. Hey, what do you think is gonna happen?

    💬 32🔄 315⭐ 420

Replies

  • 💬 0🔄 0⭐ 7
  • Oct 8, 2026, 6:32 PM

    @cwebber it also drives me crazy how parochial my fellow Americans are. Even if they did decide, at the federal level, once and for all what the status is of output in the US, that's not applicable elsewhere. And we're in an international environment where trust between national leaders is low and some sort of treaty on this seems unlikely.

    💬 3🔄 3⭐ 35
  • 💬 0🔄 3⭐ 7
  • Daveydavey_cakes@mastodon.ie
    Oct 8, 2026, 7:31 PM

    @thomasjwebb @cwebber yeah as a not-American, way too many conversations are centered on American law, like you're not allowed have an opinion if you haven't memorized the main legal system for tech companies.

    Still less annoying than when all the national governments move at the speed of pitch to enforce perfectly enforceable laws against these companies.

    💬 0🔄 0⭐ 14
  • Oct 8, 2026, 10:03 PM

    @cwebber another thought I just had is what if someone runs their agent in yolo mode (don't ask permission before running whatever tools) and it ends up "going rogue" and verbatim copying code from a repo (or a local copy of a client's proprietary code)? Sometimes the commands it runs modify the code (such as search and replace stuff or scaffolds). I'm sure someone thought of this before, but I'm not even sure that training data being sufficiently atomized is enough to prevent liability.

    💬 1🔄 0⭐ 4
  • Oct 10, 2026, 10:29 AM

    @thomasjwebb @cwebber

    I keep thinking it misguiding to say that training an LLM (or similar) on copyrighted work is a violation. It seems to me that the violation is in the *use* of the "model" to create infringing works. And the rulings' reasoning, that I've read, seems to be on the *use* being the justification for finding the violation.

    Like, memorizing someone else's song is not a violation. Performing it for others, in a way that completes with the original, is the violation.

    💬 1🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 12
  • Oct 8, 2026, 8:54 PM

    @aeva@mastodon.gamedev.place @cwebber@social.coop I'm involved in this very scene (decomps/PC ports of Nintendo games).

    This is 100% already going on. It's like every goddamn day there's another announcement by sloptubers about a "new decomp" or a "new recomp" or "a new port." And yes, of course the quality is garbage.

    And yes this does suck for the people putting in real work to have worked years on putting in the effort on a single game, only for somebody to tell Claude to slopcode the port the moment decomp hits 100% fuzzy match or whatever. Deadass been hearing people say shit like "dude I need to take my decomp private so people don't slopcode the port" and like, jeez.

    💬 2🔄 2⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 7🔄 1⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 6
  • Oct 8, 2026, 9:19 PM

    @aeva@mastodon.gamedev.place @cwebber@social.coop Was also one project that evidently knew about decompilers but didn't know about demanglers. It was all C annotated everywhere for the C++ mangled names, vtables, everything. Truly groundbreaking.

    (It should be noted that many of these decomp projects rely on symbol maps left on retail discs or other niche releases like kiosk demos. Shit dude Pikmin 1 had an entire Windows exe of the game on-disc, though it's broken)

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 5
  • Oct 9, 2026, 6:20 AM

    @pjb @cwebber @aeva It's funny how the only English word is 'unconditional' and it is itself used as a condition.

    And now I have a song stuck in my mind: ... what condition my condition was in 🎵🎶

    💬 0🔄 0⭐ 3
  • Oct 9, 2026, 12:05 PM

    @pjb Yo! That looks like the shit that comes out of COBOL/mainframe -> Java/Spring generators! Sooooo maintainable, definitely clearly and accurately defines the process.

    @cwebber @aeva

    💬 0🔄 0⭐ 1
  • Oct 9, 2026, 3:24 PM

    @pjb @cwebber @aeva from what I've seen it's like if you cycle accurate emulate the cpu and then grab that output and make some C++, so yeah it will look unreadable.

    Port probably isn't accurate, more like conversion.

    I prefer sloppy copies.

    I don't mind these, it's like static emulation with options for enhancement (like higher rez etc) and better use of CPU than pure emulation.

    💬 0🔄 0⭐ 0
  • Oct 10, 2026, 11:29 AM

    @pjb @cwebber @aeva hoooly hardcode thanks for actually posting that, and with alt! As an outsider I've heard a lot but I hadn't realized this was the fuckin state of it. Jesus christ

    💬 0🔄 0⭐ 0
  • Oct 8, 2026, 6:43 PM

    @cwebber one of the latest trends apparently is LLMs adding signatures of actual cartoonists to generated comics. I'm sure that will go over fine with everyone.

    💬 1🔄 2⭐ 18
  • 💬 0🔄 0⭐ 0
  • ChuckChuckMcManis@chaos.social
    Oct 8, 2026, 6:51 PM

    @cwebber I don't think the 3rd Circuit's reasoning is going to hold up, mostly because it would imply that natural intelligence using someone else's ideas to start their own business would be an infringement of the work. That it can be 'true' in movie scripts and 'false' in fast food stores is an issue. The *only* good thing might be to clean up US copyright law, but also unlikely.

    💬 1🔄 0⭐ 4
  • Oct 9, 2026, 12:30 AM

    @ChuckMcManis @cwebber I mean if natural intelligence uses copyrighted materials in a non-/minimally-transformative way to directly compete with the copyright holder, that's a textbook copyright violation. This isn't really any different. I agree US copyright law needs to be cleaned up though (so to for most of the world).

    💬 0🔄 0⭐ 2
  • 💬 1🔄 0⭐ 2
  • ChuckChuckMcManis@chaos.social
    Oct 9, 2026, 1:54 AM

    @be
    Coming from a place of love and support; that comment sounds like you're giving up before you've put any effort into fighting. I too can feel that sort of disillusionment but often get inspired by Carl Malamud's efforts at keeping the law public.

    @tarix29 @cwebber

    💬 0🔄 0⭐ 3
  • Oct 8, 2026, 7:16 PM

    @cwebber Problem is copyright protections will only (practically) be applied by big corps against independent creators.

    When big corps want to train on material from independent creators, they'll be given a free pass.

    💬 2🔄 1⭐ 15
  • 💬 0🔄 2⭐ 21
  • Oct 10, 2026, 3:08 AM

    @faoluin @cwebber That has already happened. The “compensation” in the Anthropic case is bullshit, and the only reason we’re getting anything is that Anthropic used a pirated copy to train its models.

    💬 0🔄 0⭐ 0
  • Oct 8, 2026, 8:02 PM

    @cwebber Video game "vibewashing" is very active in practice. (Modern Vintage Gamer did a video recently on that: youtu.be/mYZPHdKKMaI).

    A quick look on Quiver (quiverlauncher.com/) reveals quite a few recomps of Nintendo games, including big franchises like Mario and Zelda.

    That these aren't getting lawyer-bombed into oblivion is quite revealing in my opinion, as if it was somewhat in the interest of by corporations that "vibewashing" remains legal.

    [...]

    💬 3🔄 0⭐ 3
  • Oct 8, 2026, 8:09 PM

    @cwebber [...]

    For sure lots of companies have heavily pushed for using genAI for coding in the name of "efficiency", and might be afraid of all that extruded code to suddenly become an unknown liability (blaming the human coder as crumple zone will only go that far)

    [...]

    💬 2🔄 0⭐ 4
  • Oct 8, 2026, 8:10 PM

    @cwebber [...]

    I suspect that lots of big corporations are *actually relying* on chat bots to extrude license-washed version of libraries that they incorporate in their product and know that pushing against vibewashing could backfire badly.

    💬 1🔄 0⭐ 5
  • 💬 0🔄 0⭐ 0
  • Oct 10, 2026, 1:44 AM

    @dryak First time I've seen the phrase "crumple zone" used for that accountability sink position. I love it.

    💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • Oct 8, 2026, 8:38 PM

    @dryak @cwebber It may well be that they're nervous about setting a court precedent they don't like.

    They'll wait until they have a case they're 100% sure is a slam-dunk so that then they can reference that ruling in all future lawsuits.

    💬 0🔄 0⭐ 1
  • Oct 8, 2026, 9:04 PM

    @dryak@mstdn.science @cwebber@social.coop I think it's far more likely they just don't care, because emulators already exist anyways. I don't think there's been any precedent for Nintendo taking ports down as long as you need to provide the ROM yourself. Projects like Ship of Harkinian have been fine for years and widely covered in press, so I'm sure they're aware.

    Of course, that is merely an observation and I'm sure that if Nintendo did start paying lawyers to submit C&D's, everybody would just scram like ants. Who knows if something might change their mind.

    💬 0🔄 0⭐ 0
  • 💬 2🔄 0⭐ 3
  • Oct 8, 2026, 9:13 PM

    @cwebber@social.coop @dryak@mstdn.science I mean yeah, I'm aware they went after Switch emulators. But unless I missed something, I don't think they've gone after anything else yet?

    But yes, perhaps they're just waiting for the right time to
    really dunk on the scene and make an actual legal precedent. Who knows.

    EDIT: wait no I just remembered. They definitely have gone after homebrew dumping tools etc for the Switch too. Sigh.

    💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • Oct 8, 2026, 9:22 PM

    @pjb @cwebber
    Nintendo also went against Dolphin (Wii & Gamecube), Citra (3DS). And that's just the recent ones, and only emulators (we won't even mention fan-games).

    They simply don't like the concept of people enjoying stuff how they like instead of the very narrow way-to-enjoy that Nintendo wants to impose as the only authorised one. They consider such "unauthorized fun" to be a violation of their IP rights and they have the lawyer warchest to fight on that hill.

    [...]

    💬 1🔄 0⭐ 1
  • Oct 8, 2026, 9:24 PM

    @pjb @cwebber [...]

    Nintendo remaining silent in the light of AI-assisted decomp/recomp is very fihsy - as if they feared even more the impact that outlawing vibewashing could have.

    (Evermore so when you look at their "emulators for me, not for thee" stance.)

    💬 1🔄 0⭐ 1
  • Oct 8, 2026, 9:28 PM

    @dryak@mstdn.science @cwebber@social.coop I suspect the worst is yet to come as my personal observation is that the phenomenon is still very recent, and all the videos of slop ports running I can think of have been based off mostly-human-made decomps. Though maybe I haven't been paying enough attention at that.

    (Claude, take the port and the Aurora compat layer and smash them together, keep fixing crashes till I see the Wind Waker title screen)

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 8
  • 💬 0🔄 0⭐ 6
  • Oct 8, 2026, 9:15 PM

    @cwebber The US jurisprudence and political landscape is so rancid that I fully expect the eventual rulings to legalize infringement by the LLM companies, while criminalizing it for individuals.

    💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 1
  • Oct 8, 2026, 9:30 PM

    @cwebber@social.coop
    The training is one part.

    The other is the use of AI-generated code/ materials. Does that count as fair/ legal use?

    Because even if AI training on copyrighted material isn't legal, the output of that is the model - but the output of the model is a separate part of the process, and it's clear that specific groups of outputs clearly are okay and do not infringe.

    Which means that a set of criteria will need to be defined to determine if a specific use of LLM technology is legal or not.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • 💬 1🔄 7⭐ 27
  • 💬 0🔄 1⭐ 5
  • Oct 8, 2026, 9:36 PM

    @cwebber by the way that also means that pieces that have no copyright at all may not be used for llm training. At least the rule in many countries is that if you created something, put it out for free on the internet, does *not* mean one can do anything with it, such as reselling. You automatically have that copyright and anyone who wants to use it for their business must ask you.

    💬 1🔄 0⭐ 1
  • Oct 10, 2026, 1:52 AM

    @pythno @cwebber Not sure what you mean. Under US law, and that of jurisdictions that use the same or a similar system, you're right: Copyright is automatic. And merely posting something publicly is not the same as waiving that copyright. But if something does have no copyright at all (i.e. true public domain), then unless it infringes *other* rights (like trademark, personality, moral rights in countries that have them) then wouldn't that mean it can be used for anything?

    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • Oct 9, 2026, 9:39 AM

    @rachel just make sure it escaped containment and go rogue first, so you won't be personally responsible against Nintendo

    @cwebber

    💬 0🔄 0⭐ 2