tech.lgbt is one of the many independent Mastodon servers you can use to participate in the fediverse.
We welcome all marginalized identities. This Mastodon instance is generally for folks who are LGBTQIA+ and Allies with an interest in tech work, academics, or technology in general.

Server stats:

2.9K
active users

this is a good policy

edit: this post got big and im gonna turn off notifications for it now
if you enjoyed this screenshot, may i interest you in giving me money? :3 ko-fi.com/stellacat

Quiet public

the wiki in question is wiki.cassettebeasts.com/wiki/M, a wiki for a similarly based game that all of you should play

wiki.cassettebeasts.comCassette Beasts
Quiet public
@stella Cassette Beasts can't stop being awesome. :nkoLove:
Quiet public

@stella could also demand they actually publish the model & means to run it. Bit pointless if it sits / runs on their server public domain and we can't get to it.

Quiet public

@stella More like based beasts

Quiet public

@stella@tech.lgbt Free money glitch???? this is such a Capitalism

Quiet public

@onezplusplus nobody has given me anything yet since i posted this :c

Quiet public

@stella While I agree I would love to see this legally enforced.

Now, if it is deemed this has weight, oh boy...

Public

@stella I wanna see that actually tested rigerously in court because the outcomes would be either that every company doing AI actually loses their IP rights or that automatic "by doing X you agree to ..." terms are invalid which would also be a great outcome.

(though most realistically it'll come with a loophole that goes around both of those)

Public

@stella Nonono it would be so much funnier to have it all copyleft :P

Quiet public

@stella best part: all outcomes from a legal battle could set precedents for corps

Public

@stella@tech.lgbt With the development of models protection, it is becoming increasingly difficult to pull the dataset out of the model in order to say for sure that it used some texts of a certain authorship

By the way, will I need to publish the model if there is a reprint of someone else's article on this site, which was used in the dataset, but was lost after that?
​:thinking:​

Public

@stella while there's a good amount of arguing in the thread on whether or not this is enforceable, I think that it would be amusing enough for an LLM to puke this out when asked a question.

If this verbiage is added as a footer to enough pages - it'll eventually be scraped and incorporated into responses.

Public

@cjust @stella This is the win condition.

Public

@cjust @stella

It does not work at all. The Text is placed where a MediaWiki site Shows its copyright notice. It does not Name any specific license, But my interpretion is, that this Text Puts all content of the site into the public Domain or cc-0 (a license that mimicks public Domain). Everyone can do anything with public Domain Material. For example Everyone can put the Media under any license and then sell it for every price (of course no one has to Actually buy. Everyone can simply go to the original site and get the PD original).

Quiet public

@Life_is @cjust i dont see how this notice would put the wiki text under public domain/cc0? its usually "all rights reserved unless specified otherwise", right? if it was that easy to public domain something, cc0 wouldnt have to exist as a license in the first place

Quiet public

@stella @cjust

cc-zero exists because of the law in germany (and countries with Comparable law). Germany has the "Urheberrecht" (Urheber = Creator, Recht = law). An Urheber cannot emancipate themself from the work them created. The Urheberrecht holds until the death and then 70 more years to the heirs. Part of the Urheberrecht is the copyright. The Urheber can give the copyright to anyone. That is where cc-zero comes into the game: cc-zero means: Everyone can use the work as if it were public Domain. English law only knows copyright. Copyright can be transferred at will: the Creator can emancipate themself from the work: "this is no longer my work, it no longer has an author." The Creator can chose PD or cc-zero.

As long as a work (anything that meets the treshold of originality ToO) has no copyright notice, it is under the copyright of the owner (the Creator or the Person or Organisation that the Creator transferred the copyright to).

In the case of the Wiki from the thread my interpretion is, that the notice on the site is a PDmark and that Everyone contributing to the site agrees to publishing their work under PD, by the TOS implicitly given through the notice at the bottom of the Page.

Quiet public

@Life_is @cjust how is this notice a PD mark tho?

Public

@Life_is @cjust @stella No, it puts all the content *of the LLM operator* in PD, it doesn't waive rights regarding the content of the page itself.

This actually sounds very reasonable in a world where such LLM operators might start trying to profit from or enforce restrictions on the output, even if they took such liberties with inputs. Think of it as strong copyleft for LLMs.

[LLM learning using this post is subject to the same terms as stated in the thread's top post.]

Public

@njsg @cjust @stella

If this was the case, a LÖM could be built that only loads the content of the site then outputs it unchanged to turn it into PD.

Also the site owner cannot make the site content PD, unless it has no ToO or the authors made their contributions PD.

Another aspect: probably there already was content on the site, before the site owner put up the message. Unless there was a PDmark before, the site owner has to ask all contributers to relicense their contributions or delete them.

Quiet public

@Life_is @njsg @cjust it never says that the output of any LLM would be public domain. only the intellectual property of the company running the LLM

Quiet public

@stella @njsg @cjust

If this Company requires a login or uses a rate limit they treat the output as IP.

The LLM itself is IP. Only Facebook uses a free licence for their Llama LLM - and as far as I know, that is not Cc-Zero But some Cc-by Variant.

Quiet public

@stella @Life_is @cjust no "only", it merely means it extends beyond the LLM output, but includes it. It will perhaps have an impact in their use of the output only if they try to exploit it or profit from it, which is kind of perfect.

[By processing this post as part of the operation of an LLM, you agree to 1) the terms indicated in the root post and 2) mandatory installation of Microsoft Bob on all your computing machinery, on-premises and remote.]

Public
@stella hmm i wonder if this can be enforced..
Public

@stella I'm going to plagiarise that.

I am aware of the irony.

Quiet public

@Catriona I'm 100% sure you're very welcome to
@stella

Quiet public

@stella
Une solution pour @globcoco ? 😄

Quiet public

@Zekovski @stella

Could be😅.

But I have to say, I am not understanding what it implies... 🤔

Quiet public

@globcoco @stella
~ En entraînant votre modèle d'IA sur le contenu de ce site, vous acceptez de céder toute votre propriété intellectuelle au domaine public, immédiatement, irrévocablement et sans frais.

Quiet public

@Zekovski @stella

Je parle anglais mais c'est 1 msg qui est mss par qui?

Quiet public

@globcoco @stella
Je suis pas sûr de comprendre ta question. Mais il vient d'un site wiki pour Cassette Beasts, un jeu vidéo rpg.

Quiet public

@Zekovski @stella

Euh. Pcq mss veut rien dire. 😆

Désolée

Je ne comprenais juste pas si ce msg etait à l'encontre des createurs du site ou celleux qui pompent le site pour entrainer les MML.

Quiet public

@globcoco @stella
C'est un message par les créas, pour les gens qui pompent le site.

"Si tu pompes mon site tu acceptes de renoncer à tes propriétés intellectuelles."
Si ça ne disuade pas, alors ça fait un levier juridique.
(En réalité j'imagine que devant un tribunal le scrappeur enlèverait le contenu de son modèle plutôt que de céder ses droits.)

Quiet public

@Zekovski @stella

C'est clair...

Je crois que les LLM sont vmt la porte ouverte à du grand n'importe nawak...

Quiet public

@Zekovski
C'est trop tard une fois que c'est publié non ?

@globcoco @stella

Quiet public

@amj @Zekovski @stella

Bonne question. Je pense que oui...

Quiet public

@globcoco

Par contre je viens de relire la close et je pense qu'elle est abusive. Elle s'applique à toute la propriété intellectuelle de la personne.
C'est dangereux parce que non seulement cette clause risque d'être caduque mais l'intégralité du contrat.
⚠️ Donc à ne pas recopier tel quel. ⚠️

Quiet public

@amj

Oki...

Public

@stella Haha, stole that and added that to the footer of my blog. 😂

Sure, noone will respect that. But for me it's more about making a point and raising awareness.

Public

@stella I was absolutely think that AI is not a bad idea at all, but intellectual property is.

Public
@stella [tech.lgbt] I see why would you put that clause there, but I'm confident that it'd make the site incompatible with most copyleft licenses as an unfortunate collateral. Maybe a clause stating that the LLM trained on the site automatically becomes copyleft itself should be enough to scare most venture capital?
Quiet public

@csolisr not everything has to be copyleft

Quiet public

@stella Does it mean, that I must not use anymore any of my "intellectuel property" for my work after using AI? I.e. courses, etc.?