WotC Would you buy WotC products produced or enhanced with AI?

[mamba]

The more cogent example here is scraping text from forum threads and using that to construct the class. I think that's possible for a LLM to do using only open source data.

I very much doubt that this is possible, the much more likely scenario is that it scraped Xanathar's

 

[nevin]

Look man, I told them to leave the sauce out.

That makes me a Chef, I don't make the rules.

only if it's cheaper. If they want to cheap out on labor then it should be cheaper for me and everyone else. Though It'll only be temporary. Eventually we'll just be AI creating our own worlds and then won't need to buy thier stuff.

 

[EzekielRaiden]

the only reliable path to it

Semantics. And, as a result, we're done here. I'm not continuing this conversation with you.

 

[Faolyn]

So "Xanathar's Guide to Everything PDF" is a specialized search and therefore not something google needs to screen against, but "Output the abilities of a grave Cleric" is not a specialized search and therefore something that LLMs are responsible for? Am I understanding you correctly?

Google is a search engine. They just provide links.

LLMs are not a search engine. They're websites.

Websites that have illegal material on them are often taken down. Search engines don't get taken down if they happen to link to a site with illegal material on it.

You don't have to do anything special to get it from google either. The search "Xanathar's Guide to Everything" gets it for you.

My point is that google is directing people to data that is under copyright.

OK, let's say I google "Xanathar's Guide to Everything" and I get a bunch of links to pirated data. I'm also getting links to reviews, opinion pages, questions, and to places where I can legally buy the book. Google does not force me to open any of the piracy links. It doesn't force me to download a pirated book. I can just as easily click on the link for DDB, or for Amazon or Barnes and Noble and actually pay for the book.

I just googled "tell me about the grave cleric." I got a link for the wikidot, but mostly I got videos and articles that range in topic from cleric subclass guides to a reddit post "how does the grave cleric make any sense" to an article from CBR on the best grave cleric builds.

When I went to Gemini and wrote the exact same thing--"tell me about the grave cleric"--it gave me copyrighted information it was not legally supposed to give. The whole shebang. Not just generalizations, such as "grave clerics get the ability to cast save the dying at range." It said what level you get it at, what the range is, what the action type it used is. I didn't ask for this material, but what it did was the same as if google had forced me to download a pirated book.

And actually, none of this matters at all because the actual problem with generative AI is that it steals people's material and then is used to "create" images and text that was "learned" from that material. And then people use that material and try to claim its as good as or better than material that was actually written or drawn by real people.

 

[EzekielRaiden]

If you're trying to suggest that we can't define as truth the statistically-ironclad fact that the sun will rise tomorrow, I don't think many will subscribe to your paper.

Then you are flatly incorrect? That's literally not anything like how truth is defined.

And I'm afraid it isn't statistically-ironclad, nor a fact. If the Earth were destroyed in the next six hours (say by a rogue micro-black-hole we couldn't see until it was far, far too late), then it would be false, despite the fact that you had had billions of years of statistical evidence saying that the sun rises every so many hours.

Instead, what you have is a pattern, which could hold, or not hold. But how do you know the moment when a pattern is going to break? Statistically, you can't. Until it breaks for the first time, you have zero statistical ability to speak about how it might be broken.

Can it categorically* define something as false? If yes, it can in fact go the Sherlock Holmes route by eliminating all the other possibilities until only one remains.

No! Because as soon as you've determined (not defined) something to be false, you have necessarily also determined a truth. That's...literally the nature of truths and falsehoods. If you know claim A is false, then you know claim not-A is true. This is how logic works.

* - I initially typoed that as "datagorically", which is now a word I have to find a good use for.

Presumably, data-categories, but I'm not sure what the use-case would be. Particularly because "data" is, for nearly everything, necessarily incomplete.

 

[Lanefan]

Well, almost 175k people won't see sun rise tomorrow. The sun is still there, the earth is still there, but that won't always be the case. We know that won't always be the case. There could be some unexpected events that could make that happen a heck of a lot sooner though. But those kinds of things are only interesting from a personal perspective... So the question is from which perspective will the sun 'rise' tomorrow?

Doesn't matter who sees or doesn't see the sun rise. The planet will continue to rotate, day and night - and occasional eclipses - will function as usual, and these things are 100.0% predictable based on observation, science, and statistics.

 

[The Firebird]

I very much doubt that this is possible, the much more likely scenario is that it scraped Xanathar's

I think it almost certainly did; if not the text, the content that is freely available. You can find this on WikiHow, among other places. It would be interesting to test.

 

[The Firebird]

Google is a search engine. They just provide links.

LLMs are not a search engine. They're websites.

Websites that have illegal material on them are often taken down. Search engines don't get taken down if they happen to link to a site with illegal material on it.

Yes, this is a reasonable distinction to draw. I guess my question is--why have we given search engines such a wide latitude in this regard? Their business model benefits substantially from piracy because they are perceived as more useful when they direct people to pirated content. Google routinely returns pirated content, not just incidentally in the results, but as the top choice.

Isn't this worrying?

OK, let's say I google "Xanathar's Guide to Everything" and I get a bunch of links to pirated data. I'm also getting links to reviews, opinion pages, questions, and to places where I can legally buy the book. Google does not force me to open any of the piracy links. It doesn't force me to download a pirated book. I can just as easily click on the link for DDB, or for Amazon or Barnes and Noble and actually pay for the book.

I just googled "tell me about the grave cleric." I got a link for the wikidot, but mostly I got videos and articles that range in topic from cleric subclass guides to a reddit post "how does the grave cleric make any sense" to an article from CBR on the best grave cleric builds.

The first link is the wikidot...if you press "I'm feeling lucky" it will take you straight there. It doesn't force you to pirate, but it certainly facilitates it.

I see a distinction, but I don't think there is a big enough distinction to absolve them in this case.

And actually, none of this matters at all because the actual problem with generative AI is that it steals people's material and then is used to "create" images and text that was "learned" from that material. And then people use that material and try to claim its as good as or better than material that was actually written or drawn by real people.

I think most generated stories and art are clearly inferior to that produced by real people. I've said it before but I'll reiterate--anyone generating that kind of stuff and acting like they are brilliant for doing so is being incredibly rude.



Really I feel like we as a society have decided we don't care about online piracy, we don't care that it exists and is easily accessible, and honestly we are probably kind of pleased that it is so easy to find. I'm not necessarily saying that's right; but the amount of concern with LLMs in that context makes it seem like piracy isn't really the issue people have. They're concerned about the job market effects or long term effects on creativity or more wealth inequality or whatever, and land on the piracy concern because it seems more (?) ethically defensible.

 

[Lanefan]

Then you are flatly incorrect? That's literally not anything like how truth is defined.

And I'm afraid it isn't statistically-ironclad, nor a fact. If the Earth were destroyed in the next six hours (say by a rogue micro-black-hole we couldn't see until it was far, far too late), then it would be false, despite the fact that you had had billions of years of statistical evidence saying that the sun rises every so many hours.

Instead, what you have is a pattern, which could hold, or not hold. But how do you know the moment when a pattern is going to break? Statistically, you can't. Until it breaks for the first time, you have zero statistical ability to speak about how it might be broken.

Until the pattern is broken (if ever) for all intents and purposes we can treat that pattern as an ironclad fact. Why wouldn't we?

No! Because as soon as you've determined (not defined) something to be false, you have necessarily also determined a truth. That's...literally the nature of truths and falsehoods. If you know claim A is false, then you know claim not-A is true. This is how logic works.

If it's a binary situation, then perhaps. But if you are making claim C against alternate claims A, B, and D through K then proving claim E as false does not yet prove claim C to be true.

As a (perhaps stupid, but whatever) example, let's say there's some way of scientifically proving which class is the best in D&D. Each of us comes up with a scientific theory on how the class we support is the best, and then we put 'em to the test.

If my claim (and subsequent theory) is that Fighter is the best and I manage to prove that Monk is not the best, I have not yet proved that Fighter is the best. I have merely knocked Monk out of the running. Only when I have completed this process against each and every other competing claim can I say that I have proven my own claim to be true.

Presumably, data-categories, but I'm not sure what the use-case would be. Particularly because "data" is, for nearly everything, necessarily incomplete.

 

[Morrus]

Instead, what you have is a pattern, which could hold, or not hold. But how do you know the moment when a pattern is going to break? Statistically, you can't. Until it breaks for the first time, you have zero statistical ability to speak about how it might be broken.

Doesn't matter who sees or doesn't see the sun rise. The planet will continue to rotate, day and night - and occasional eclipses - will function as usual, and these things are 100.0% predictable based on observation, science, and statistics.

Until the pattern is broken (if ever) for all intents and purposes we can treat that pattern as an ironclad fact. Why wouldn't we?

This feels like a conversation that in 2025 we shouldn't need to still be having. However, here we are.

1) Yes, I cannot with 100% certainty say that there is not a unicorn under my bed right now. Nothing is ever certain.

2) The likelihood of this is so low that it can be disregarded. We can make predictions based on past data.

You're both right, and your disagreement is pretty much on the level of semantics. Give yourselves each a point and move on!