Skip Navigation

Posts
8
Comments
1826
Joined
2 yr. ago

  • ever tried befunge?

  • dust bunnies.

  • something something solar roadways

  • i mean i agree about all the stuff about plagiarism, concentration of power, usage of limited electricity and water and land, mental effects etc. i've expressed as much before. the thing i think is interesting to discuss is that, given that there's already a generation of old pre-trained models, and that there's no fossil fuel generation in my country, and that i'm doing this rather than playing a game, which uses the same amount of power, and that i'm not sharing results, just making personal scrapbook stuff... is that more immoral than just pirating movies or music?

    that's not supposed to be a trick question, i genuinely don't know. i've always been of the opinion that media empires shouldn't make money from me but artists absolutely should, and that's why i generally pay indie artists but not big labels. here, i use three- or four-year-old model files which i've not paid for, on my own hardware that's completely offline so no data is provided to the big companies. i always recommend against using online commercial models and i refuse to do it myself because of the things you mentioned, but... is what i do the same?

  • surely a useful idiot is one who furthers the message and aids in its spreading?

  • i also think that. and i love having that discussion with people. like, remember the google robot that drew dogs on every picture you fed it? deepdream, i think it was called? was that evil? because it's the same tech, just trained on... dogs.

  • i don't think you should start calling third parties useful idiots, i'm the one who started this.

  • correct. though stable diffusion was initially trained on LAION5, which is a free and open dataset compiled for scientific research which in itself had copyrighted material in it. that didn't come out until after the model was out there because it didn't matter to anyone before. scientists had used it for deep learning tasks for years at that point.

    and the source images i take i also do without payment or copyright acknowledgement. as is the shows i watch on my jellyfin server, and some of the computer games i play, and the music i used to record off of the radio onto cassettes, and the comic characters i used to cut out of magazines as a child to make my own comics with.

    i'm not saying those things are all equivalent. i'm saying that this problem isn't new. i'll stand up and defend small artists against big corporations any day, and i'll gladly pay people for their time. my patreon bill alone is proof of that. i've canceled netflix and spotify over their unfair treatment of artists. i've written to my members of parliament, both local and in the eu, about how the copyright system is broken, how the slogan "information wants to be free" doesn't automatically mean that meta can leech eighty terabytes of copyrighted material or that the us government can use any work they want in their propaganda material without paying. i've driven discussion about rightsholders and the unfairness of payouts to the little guy both at work and during my off time. i've voted against corporate control of media for years. i've voted pirate. i've voted socialist. i've voted green.

    but sometimes i just want to dick around.

    believe you me i've thought about this.

  • hi felix

  • i don't really want to class it as such in the current climate.

  • well yeah, sure. not denying that. but i don't do it for people to see it. i think of it a bit like cutting stuff out of magazines and gluing it together. it's remix culture.

    i'm not a creator in any way, i'm a consumer. i just like blending things together.

  • i agree with you, but i'll give you a counterexample.

    i generate ai images for me. i use only local models and i never share it publically because the output is not really the point, the process is. here is my process:

    i come up with a concept, usually a person or a scene. i then take random images from the internet, cut out the parts i think fit together, and add them as layers in a client called "invoke ai". if needed i color match the parts in krita first. then i describe the scene i've made in a prompt, adding the normal positive and negative keywords to steer generation. i also pull the "blur image" ratio down to 40%.

    the model then makes an image with my digital scrapbook as a base, melting together disjunct elements into a scene. invoke then allows me to move all the elements around and regenerate the scene, or select a few specific elements to regenerate, or paint on top of the scene and generate something new from that, or select part of the scene and change the prompt for that area. it's a fun little game, and it feels like collaborating with people who know lighting and perspective better than i do.

    most time i've spent doing this is half a day, just iterating and tweaking and filling in details. i'm in no way an artist, i'm playing around. it's basically as resource intensive as playing a video game, and i'm not one to share gameplay footage either.

  • i've had to deal with situations like that before, not specifically because of gpl but because of international regulations. one of my customers was a digital id provider, and they had one of those super-accurate timekeeping/cryptography servers they needed to move from lithuania to sweden. because of laws surrounding encryption of personal information, the server would count as compromised if there was ever a single second where it was left without supervision. so they had two of their people drive non-stop through poland, germany, denmark and half of sweden with the server on a ups in the back seat.

  • well then. my argument is moot.

  • sounds like it says that to me. "we can't send you the source over the internet because of security reasons so you need to pay us for a plane ticket so one of our representatives can give you a cd directly" is evil and stupid but completely reasonable in a legal sense.

  • i'm using "on top" rather flippantly here, since orca is AGPL. but bambu may also have separate code running on the machines that is not agpl.

  • yeah, this is the interesting one. also this one and he ones below it.

    basically, bambu has the right to refuse handing out gpled code. that's part of their freedom to distribute as they see fit. however, they can only exercise that freedom on people they haven't given the binaries. so if you decompile or download all the sources, and you don't own one of their printers, you are also violating the license.

    now if rossman owns a bambu printer, and he has gotten the sources from that printer or directly from bambu, they can't do shit. otherwise there is wiggle room.

  • yeah, a lot of times. i had software licencing responsibilities for a product for a while, and they really didn't want gpl stuff in there for the reasons stated in the thread.

    the interesting wrinkle in this situation is: if rossman is distributing all of bambu's code, including the account stuff that far as i can tell is another codebase, and he hasn't gotten it from his own printer (which wouldn't have the sources) then no "agreement" has taken place. if bambu is told to distribute sources and they say "no", they're in the wrong. but if they haven't said anything, then technically rossman is stealing the code. it would most likely get thrown out, but the case can be made.