That explains why so often the titles I get are utter shit, and many of the articles I've been getting lately seem really incoherent.

//

*Copyright infringing. If it were merely being copyrighted that made distributing it bad then you wouldn'ty be able to share that Creative Commons licensed song you or someone else made, since it'd still be a "copyrighted work" due to copyright in the US being automatic upon an eligible work being put into a fixed tangible medium.
/pedanticMode

Not copyrightable if you just take what is generated, but if you modify it enough, it can absolutely IIRC - making it not useless per-se (depending on how much human touch is needed).

I mean, it isn't just companies training models, there are open source models, people contributing their own training models for various purposes.

IMO that's a hell of a false equivalence, on many levels.

I mean, that wouldn't be the boss' choice - that is decided in the courts, and whether it is a fair use or not is still up in the air. (This would also age like milk if it was ruled to be a fair use).

In the simplest sense, isn't that what everyone does in some way or form?

Still, depending on the game, deducable - granted, it'd be a right pain in the ass, but if the TF2 botting problem is anything to go by, creating accounts to test each thing individually would be easy.

Perhaps this is a really stupid question, but would those who are working on cheats be able to determine that anyways? It seems like something that could be deduced rather easily.

The training set, or data set used to make images? (since both are sets that exist, but are different sets for different purposes. The latter ... that'd be absolutely false if everything went to plan (including avoiding overfitting/overtraining), just look at how much smaller the actual generation data set is compared to the training data, for instance.

What I mean is, unlike closed source models, I am struggling to see how you'd enforce such things on them.

Since in many countries anything eligible for copyright is considered copyrighted upon creation - which includes works where explicit permission was given, or the work was licensed in a way where such use is allowed (various forms of creative commons licensing).

This is what bugs me about the way people address copyright in these debates; they make it about copyright status, saying things like "train without copyrighted works," or "ban copyrighted works being used," ignoring that because of how copyright works, that'd kill a lot of allowable uses (and that's without touching the debate over if such usages w/o consent constitues fair use or not).
Basically, it incorrectly conflates copyright status, with licensing status - which is especially annoying given how fucked people's understanding of copyright tends to be (no thanks, IN PART, to the RIAA, MPAA, etc in the 00s), being careless in the way I pointed out just exaserbates that misunderstanding of copyright to our detriment.

Registered and tracked how, and what about things like open source models?

Got a source for that?

As with a lot of "X ruins Y for Z" claims, IMO it relies on the assumption there is only one variable involved or one aim of blame, faultily.

I mean, if something happens, and a party overcorrects afecting people (when it is arguably avoidable), in general I mean, shouldn't the overcorrecting party hold part of the blame?