trappedslider
Legend
The ideal outcome according to some is that it becomes so prohibitively expensive that no one does it.
Disney sues Midjourney
- Thread starter overgeeked
- Start date
Look, I don't know what to tell you if you can't imagine how programs that respond to and interact with people in thier own frame of reference are powerful and useful. We have literal decades of sci-fi that highlight the ability to talk to a computer like a person and get a cogent response. This has been the goal for a long time. Just because YOU can't imagine a use beyond crappy art doesn't devalue the technology as a whole.
MoonSong
Rules-lawyering drama queen but not a munchkin
More like Gargantuan or Colossal amount n_n.
overgeeked
Open-World Sandbox
We always have a choice. Just because it’s hard, doesn’t remove the fact.
trappedslider
Legend
In this instance/case, they haven't made a choice that people dislike (yet), but they have made many many choices that folks have disliked.
The Firebird
Commoner
Search, research, programming, and tutoring are the best use cases so far, ime. LLMs do things that are useful and unimaginable, or would have required substantially more effort, prior. I haven't played with the art ones but I think it is easy to see uses.
If you restrict yourself to examples where the technology is used clumsily or for frivolous reasons, that's on you.
Building it ethically is easy. License the content if you don't own it. Pay your darned royalties, and the basic ethical consideration goes away.
kigmatzomat
Legend
I'm not surprised. Disney has been behind copyright extensions in the US forever.
Just as a vweu superficial copyright overview /"what-if" exemplars including rpg impact ( disclaimer: IANAL, YMMV, etc)
1790 copyright -14yrs that can be extended one time
The public domain had most things pre-2011(>14yro and everything pre-1997 (>28yro). In that world pretty much everything d&d prior to 3e would be public domain, as would the early editions of GammaWorld, Gurps, Hero, Shadowrun, Traveler, WoD, etc along with all the artwork. The old-school world could reprint and/or modify actual "old school editions" legally.
LLMs could be built on a "current-ish" public domain data set (on par with the "Random Grandma" in keeping abreast of current events) and then enhanced with copyrighted works by people who own/license it to be more specifically informed.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters newer than Hunchback of Notre Dame or Toy Story 1 , you absolutely would be sued into oblivion when Dusney notices you.
1976 authors life+50yrs/75yrs work for hire
This is a world where maybe the 1974 first edition d&d was public domain if it was a work for hire, otherwise that would be protected until 2059 (Dave Arneson's death in 2009 +50yrs). No other RPGs would be public domain.
LLMs based on public domain would be limited to data from the 1950s, could generate WW2-era and prior artwork, would know about Mickey Mouse and the Loony Tunes, and would be capable of doing customer service work (augmented appropriately with current product info) but not "search", would have issues with modern terms (i.e. cell phone, laptop, tablet) and would likely be incredibly racist, sexist, and very anti-communist. Programming assistance would be custom per language/compiler or need to be GPL-based.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters released after Cinderella (1950), you absolutely would be sued into oblivion when Disney notices you.
We currently live in the "life+70yr / work for hire ~95yr" universe. Meaning 1930s era. LLMs trained on public domain works should sound very outdated and say "23-skidoo!". Betty Boop would be available, I think.
So if an LLM doesn't sound like a mobster or a WWI news reel, if it knows what a jet plane is, that rock & roll exists, or it assumes that "non-whites" can go anywhere in America a "white" person can go, its almost certainly trained on copyrighted works without permission.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters, you absolutely would be sued into oblivion when Dosney notices you.
Just as a vweu superficial copyright overview /"what-if" exemplars including rpg impact ( disclaimer: IANAL, YMMV, etc)
1790 copyright -14yrs that can be extended one time
The public domain had most things pre-2011(>14yro and everything pre-1997 (>28yro). In that world pretty much everything d&d prior to 3e would be public domain, as would the early editions of GammaWorld, Gurps, Hero, Shadowrun, Traveler, WoD, etc along with all the artwork. The old-school world could reprint and/or modify actual "old school editions" legally.
LLMs could be built on a "current-ish" public domain data set (on par with the "Random Grandma" in keeping abreast of current events) and then enhanced with copyrighted works by people who own/license it to be more specifically informed.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters newer than Hunchback of Notre Dame or Toy Story 1 , you absolutely would be sued into oblivion when Dusney notices you.
1976 authors life+50yrs/75yrs work for hire
This is a world where maybe the 1974 first edition d&d was public domain if it was a work for hire, otherwise that would be protected until 2059 (Dave Arneson's death in 2009 +50yrs). No other RPGs would be public domain.
LLMs based on public domain would be limited to data from the 1950s, could generate WW2-era and prior artwork, would know about Mickey Mouse and the Loony Tunes, and would be capable of doing customer service work (augmented appropriately with current product info) but not "search", would have issues with modern terms (i.e. cell phone, laptop, tablet) and would likely be incredibly racist, sexist, and very anti-communist. Programming assistance would be custom per language/compiler or need to be GPL-based.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters released after Cinderella (1950), you absolutely would be sued into oblivion when Disney notices you.
We currently live in the "life+70yr / work for hire ~95yr" universe. Meaning 1930s era. LLMs trained on public domain works should sound very outdated and say "23-skidoo!". Betty Boop would be available, I think.
So if an LLM doesn't sound like a mobster or a WWI news reel, if it knows what a jet plane is, that rock & roll exists, or it assumes that "non-whites" can go anywhere in America a "white" person can go, its almost certainly trained on copyrighted works without permission.
And if you, as an individual, tried to make a service where people paid you money to make rip-off art of Disney characters, you absolutely would be sued into oblivion when Dosney notices you.
Those are important aspects, but I think it is unhelpful to call it "easy." Building a system for identifying, contacting, negotiating with and ultimately paying copyright holders would certainly not be "easy." Determining how royalties would work would be equally complex: what portion of the training data from any given source contributed to any given output is a hard question.
Part of the problem, of course,is that these are tech firms racing to get more and more VC money (and maybe some actual profits down the line). Tech firms are not known for their ethical (or in some cases, legal) business practices. Publicly funded and accountable research in universities with government oversight would probably have yielded more but more slowly -- which, with AI, is probably something we really want.
At this point, lawsuits like Disney's will not "end AI." Based on worth, OpenAI could just buy Disney and be done with it.
As an actual example: many years ago, back before the term "generative AI" was coined, I personally did potential physics doctoral thesis research on using the technology to simulate the data that comes out of particle accelerators, to help design and tune their detectors and data processing software.
No ethical issues there - the training data would be the output of previous accelerators, not from humans depending on copyright to make a living off it.
Similar Threads
