How well can a dedicated RPG GenAI perform?

[pemerton]

Research and transformative use are both covered under fair use. So by any reasonable interpretation this isn't even illegal. The only way this gats banned is if some corrupt judge maliciously misinterprets the law in order to prop up an obsolete industry.

Is this a professional legal opinion? Or conjecture?

I'm not a very strong IP lawyer, but my first thought on reading @Gorgon Zee's OP was that the process adopted seemed to involve copyright violations.

 

[Cergorach]

I'm not a very strong IP lawyer, but my first thought on reading @Gorgon Zee's OP was that the process adopted seemed to involve copyright violations.

How exactly? Copyright is about distribution as far as I understand it.

Let's not start the discussion on whether or not a public trained LLM model is copyright theft or not, as that's not relevant here if the LMM model is not public.

The assumption is that what's used in the model now 'global' ChatGPT is trained on. But that's only if you have that turned on, you can turn that off (opt out) in ChatGPT. Then it becomes that you train only your own 'LLM', that only you can use. Sure, you're not doing it all on your own hardware (but you could), but it's not copyright infringement when you buy the pdf and store it on cloud storage only you can access. It of course is a totally different issue when you make that available to others to read.

Now, posting the results here might be considered copyright infringement, but how is this different from every other related ENworld thread that discusses an IP? Like D&D or LotR...

The issue is imho more that people make different assumptions on what they read or heard somewhere and often the research they should do into the subject is thoroughly lacking. On both sides of the conversation. Using the wrong terms, doesn't help either: 'copyright theft' does not exist, it's 'copyright infringement'. An LLM doesn't 'think', it 'calculates', and it certainly doesn't know what a 'fact' is (not that many humans do either)...

 

[velkymx]

I've got mine working pretty good on a Open Source creative commons version.

Introducing the Cresthaven RPG Rules Lawyer GPT: Your Ultimate Game Companion

Whether you’re a game master or a player, this powerful tool simplifies complex rule interpretations, helps resolve disputes, and ensures your game flows smoothly. Instantly clarify mechanics…

www.cresthavenrpg.com

 

[pemerton]

How exactly? Copyright is about distribution as far as I understand it.

Copyright is about copying, and related reproductions/transformations.

The OP seemed to me to describe quite a bit of copying and consequent reproduction.

EDIT: a non-expert can't reconstruct the law of copyright just by reading a statute; but the statute does provide a starting point.

Here is the relevant US law, that sets out (again, solely as a starting point) the exclusive rights enjoyed by a copyright holder: 17 U.S. Code § 106 - Exclusive rights in copyrighted works

 

[Cergorach]

@Gorgon Zee How happy are you with the results?

I think it's interesting what you've done, but I think making/configuring this is way beyond the skill set of the average user and quite time intensive. Imagine doing this with the whole library of all Forgotten Realms source books and adventures over the past ~35 years...

I've been looking at MS Copilot for Office 365, which according to the MS marketing machine allows you to include documents (for reference) that you have access to (in your tenant) to your LLM responses. How well that works I don't know yet (as MS often sells you a product that isn't doing what it should), but that sounds like a nice user friendly shell that does something similar to what you did here. Any experience with that product or something similar that makes this more user friendly, less time intensive?

Copyright is about copying, and related reproductions/transformations.

The OP seemed to me to describe quite a bit of copying and consequent reproduction.

If that's the case, we make a copy of the copyrighted information every time we download it. When we move it from one disk to another, we first make a copy and then 'forget' where the previous copy was located (it's still there, just not yet overwritten). We apparently have the right to make copies, it's kind of necessary with how computers work...

Also keep in mind that US copyright isn't globally enforced. In the Netherlands I have a legal right to make copies for personal use for example. International copyright laws have local implementations, that can be very different. There have also been quite a few changes and interpretations of those copyright laws over the past 40 years (at least here in the Netherlands).

How is making a copy on disk any different then making a transformed copy on another disk (LLM)? With a pdf you can make embeded indexes, which essentially takes the content of the pdf, transforms it so it's better searchable and embeds it in the document again (changes the file). What about catalogs, which is the same feature, but for multiple pdf documents. Something I already did ~20 years ago with all the OCRed official 2E D&D pdfs. Which was a great tool for preparing adventures (search for <name> and it would give all the results from all the included books with direct links to the mentions).

What about a blind person that uses a text to speech option? How do you think those things work? A copy of the work is loaded and then transformed, there's been some AI usage in there for years... Again, depending on the country you're in, that's a right and duty to make 'works' accessible by the disabled.

There are many laws/rules that are constantly being broken and not being stopped, fined or punished. ChatGPT has been publicly available for almost two years now. Big companies like MS and Apple now use it/sell it. LLMs are a $6+ billion industry in 2024... And as far as I know (haven't been actively following this for a couple of months now) still no legal consequences. So you can yell "This is illegal!" until you're green and yellow, when no one is enforcing that, it's just shouting into the wind.

 

[pemerton]

If that's the case, we make a copy of the copyrighted information every time we download it. When we move it from one disk to another, we first make a copy and then 'forget' where the previous copy was located (it's still there, just not yet overwritten). We apparently have the right to make copies, it's kind of necessary with how computers work...

The law of copyright, in its application to software products, has regard to these functional aspects of computing. Likewise the licences that software vendors grant to their consumers.

But what is described in the OP doesn't, to me, seem to fall within this sort of category of use.

Also keep in mind that US copyright isn't globally enforced. In the Netherlands I have a legal right to make copies for personal use for example. International copyright laws have local implementations, that can be very different. There have also been quite a few changes and interpretations of those copyright laws over the past 40 years (at least here in the Netherlands).

I'm aware of this. My own jurisdiction is Australia.

But I believe the OP is in the US.

And even in the the context of the Netherlands, I would want to hear an expert opinion on whether or not what the OP describes falls within permitted use before forming an opinion on its lawfulness.

What about a blind person that uses a text to speech option? How do you think those things work? A copy of the work is loaded and then transformed, there's been some AI usage in there for years... Again, depending on the country you're in, that's a right and duty to make 'works' accessible by the disabled.

I haven't reviewed the US law for the details of this sort of thing, as I'm not familiar enough with the US statute. But here is the relevant Australian law (Copyright Act 1968 (Cth)):

113E Fair dealing for purpose of access by persons with a disability

(1) A fair dealing with copyright material does not infringe copyright in the material if the dealing is for the purpose of one or more persons with a disability having access to copyright material (whether the dealing is by any of those persons or by another person).

(2) The matters to which regard must be had, in determining whether the dealing is a fair dealing for the purposes of this section, include the following matters:

(a) the purpose and character of the dealing;

(b) the nature of the copyright material;

(c) the effect of the dealing upon the potential market for, or value of, the material;

(d) if only part of the material is dealt with—the amount and substantiality of the part dealt with, taken in relation to the whole material.​

With a pdf you can make embeded indexes, which essentially takes the content of the pdf, transforms it so it's better searchable and embeds it in the document again (changes the file). What about catalogs, which is the same feature, but for multiple pdf documents. Something I already did ~20 years ago with all the OCRed official 2E D&D pdfs. Which was a great tool for preparing adventures (search for <name> and it would give all the results from all the included books with direct links to the mentions).

I don't know the nature of the licence that was granted to you when you acquired those PDFs. And I'm not an expert in IP law by any means. So I'm not going to offer any conjecture as to whether or not what you did is lawful.

So you can yell "This is illegal!" until you're green and yellow, when no one is enforcing that, it's just shouting into the wind.

Are you arguing that what the OP did is not a copyright violation? Or that it will not be pursued by the copyright holder? Or something else.

I'm well aware that plenty of people engage in unlawful copying (unlawful in the sense of *violating a copyright holder's rights in their IP). The mere fact that something is done, even commonly done, doesn't in itself establish that it is lawful.

My post, to which you responded, was this:

I'm not a very strong IP lawyer, but my first thought on reading @Gorgon Zee's OP was that the process adopted seemed to involve copyright violations.

Are you saying that I am wrong, or not?

EDIT: Out of an irrational sense of completeness I decided to look at the US statute. The provision exceptig reproduction for use by blind people from the general law of copyright is 17 U.S. Code § 121 - Limitations on exclusive rights: Reproduction for blind or other people with disabilities

The key provision is this:

(a) Notwithstanding the provisions of section 106, it is not an infringement of copyright for an authorized entity to reproduce or to distribute in the United States copies or phonorecords of a previously published literary work or of a previously published musical work that has been fixed in the form of text or notation if such copies or phonorecords are reproduced or distributed in accessible formats exclusively for use by eligible persons.​

The rest of the section is mostly about ensuring that this exception does not permit the copies to be used for any purpose beyond that for which the exception applies.

 

[Bohandas]

Is this a professional legal opinion? Or conjecture?

I'm not a very strong IP lawyer, but my first thought on reading @Gorgon Zee's OP was that the process adopted seemed to involve copyright violations.

Conjecture.

 

[Bohandas]

In the unlikely event that it comes about that it is ruled to be illegal I very much hope that such a ruling will lead to a backlash against copyright law.

 

[Gorgon Zee]

Copyright is about copying, and related reproductions/transformations.
The OP seemed to me to describe quite a bit of copying and consequent reproduction.

EDIT: a non-expert can't reconstruct the law of copyright just by reading a statute; but the statute does provide a starting point.

Here is the relevant US law, that sets out (again, solely as a starting point) the exclusive rights enjoyed by a copyright holder: 17 U.S. Code § 106 - Exclusive rights in copyrighted works

Notes:

• I am not a lawyer, but have worked with lawyers quite a bit in IP law, but mostly concerned with patent law. My opinions are my own and should not be relied on or treated as any form of authority.
• I am happy to be wrong here. It's not like I'm basing anything major on this and I'm more interested in the theory of this process than the actual results.
• If anyone with a material interest in my project objects, I'll just stop and delete everything
• Copyright law is not core to the thrust of this thread, so I'm not going to spend a ton of time discussing it in this thread. But it is related!

So although 106 is a good start, I would consider my use of the PDF as a Fair Use. Section 107 is helpful here:

Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—

1. the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
2. the nature of the copyrighted work
3. the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
4. the effect of the use upon the potential market for or value of the copyrighted work.

"Fair Use" is not something that you can define hard rules about -- it gets argued a lot in courts -- and 107 above shows that by not defining rules, but just saying what factors can be taken into account. Essentially, if you want to definitively know if your use is fair use, you have to go to court.

In my case, my reasoning for why I think my use is "fair" as per the code is as follows:

• I am making a copy for scholarship and research purposes. Specifically the copy I use is an index on the work allowing it to be combined in new forms.
• The purpose is for nonprofit purposes. Not sure if it's "educational". I mean, it educates me, but as a layman that feels a stretch.
• The nature of the copyright work is as a reference material, with the expectation that people will want to refer to sections. The work provides an index and so the creation of a better index for research purposes fits within the work's design.
• The fact that the entire work is being indexed is a strike against this use.
• There is no apparent negative effect on the potential market.

In court cases I have seen reviewed, the courts seem to weight the personal utility of making a copy against the damage to the copyright owner. In my case, given that I am pretty confident there is no damage to the owner, the burden of proof of fair use should be pretty light.

As an aside, I should point out that I could actually replicate my system without needing to copy the work. My vector database (the copy) stores text fragments (copies) together with their indexing. Instead I could store references to the fragments and so avoid any copying. It would be a pain though...

Anyway, that's my reasoning. Not a lawyer and very aware that "fair use" is used often by bad actors, but I do honestly think my use qualifies. But since this is a matter of opinion rather than clear violation, if you think otherwise, fair enough!

 

[pemerton]

Not a lawyer and very aware that "fair use" is used often by bad actors, but I do honestly think my use qualifies. But since this is a matter of opinion rather than clear violation, if you think otherwise, fair enough!

Well, I'm not a practising lawyer, I'm an academic lawyer and as I've already posted I'm not an IP expert, and my home jurisdiction is not the US. So I'm not going to contradict your intuition, though I'm not going to say that I share it either.