AI in PbP

This is a Burden of Proof argument, which is a logical fallacy. "How do you know they don't use it for training models," is a way of shifting the burden of proof from the claimant (you) to the 'opponent' (anyone who disagrees). NotebookLLM says it does not use your inputs for training unless you opt in, or if you request feedback (thumbs up or down). So no, I don't believe that anyone using it for notes needs to notify anyone. If evidence is provided that NotebookLLM does, in fact, use your inputs for training, then I would happily change answer.

I just want to say that I appreciate when people put disclaimers about their AI usage. I certainly have no interest in joining a game which uses generative AI tools for anything, notes or not, so users being open about it saves me the heartache. Not because they'll use my work for training - it's going to be stolen one way or another even without them, - but because I have no desire to support the industry even indirectly. Which is already hard to avoid as is.

(Also tech companies proved morally bankrupt time and time again. By signing most ToS you already agree to forfeit your personal information, your derivative works and your right to sue them, so I doubt there is not a fine print somewhere about datasets too. But it's probably much less hassle to steal publicly available stuff on the web than from their users, so I won't actually bet on them using user input for training. Unless it's the good old tiny checkmark about opting in that's on by default.)

Unfortunately, the training thing by itself is just a new way of stealing stuff on the internet. With anything public, there is no guarantee that someone somewhere isn't using it in ways you'd rather they don't.

But this isn't a court of law, or scientific review, we're talking about casual relationships and trust. You can say it's a logical fallacy, but... that discounts the concern of someone with whom, in theory, you're trying to build a relationship and trust. (Your prospective collaborators in storytelling.)

Does that posture achieve that goal? (Of relationship and trust building?) Would it not be better to show that you understand your collaborators concern and will be responsive to concerns? Isn't that how relationships and trust are built?

If there's nothing to worry about and everything's ok about it, what's the problem with letting people know that you're using AI in any capacity?

Like, to you.

What do you tell yourself, if you believe it's ok, that makes you think you shouldn't or don't have to tell people about your use?

Shouldn't at the very least, under such an understanding, it be a completely neutral or benign disclosure? If not something to be proud of if one sees it as a smart life hack?

Why hide/not-disclose it?

And tbf this is a line of questioning I've wondered about from seeing people post stuff on social media, long before this discussion. I'm not even necessarily directing the question at you cowleyc, but rather anyone that falls within this ideological niche on the topic.

The question is why would I disclose? Under the assumption that NotebookLLM does not use data for training (and no evidence has been provided otherwise), then it is another harmless* tool much like spell check, random generators, or the dice roller in GP. This isn't something like a safety tool where you're disclosing to prevent harm. Assuming that someone is hiding usage by not disclosing is setting a dangerous precedent.

* Due to the RAG model of Notebook LLM, water usage is minimal, according to Google. Unless you're enriching using external sources, like with the audio features. This is a disclosure that I have used Google.

Now, for the record, I don't put my GP games into NotebookLLM. I will, however, often use it for rulebooks and settings to parse out thick tomes. In games where I advertise no AI use, however, that also applies to NotebookLLM.

I get the notion, but it does sound kinda.. defeatist.
I think a lot of writers would find (and already found and sued) their work used without their knowledge or permission in LLM training harmful.
I imagine NotebookLLM is not the only tool of that kind on the market. What about the others? Assuming we believe the makers of these models to keep their word (and not change their terms later on on the down-low).
You read ToS / license of the tools you use and disclose upfront that you used them?

It is absolutely defeatist, and cynical. But I found resigning myself to the idea makes it easier to swallow when I do discover things like that happening. If you expect it to a occur, it is more palatable (even if not less unpleasant), that kind of thinking? Plus it's a good reminder not to post things on the internet which you'd rather keep private.

...I do miss the days when I could just share art or writing publicly without that fear. At least when a human steals something directly, it's more obvious. :D But I still harbor the idealistic notion that eventually ethics will catch up with the technological progress. Even if it does look... doubtful, at times. Or maybe the actual, genuine AI appears, and then it'll do a better job regulating stuff.

---

Also more on the current topic, I think AI disclosure is good in the same way disclosing any divisive topic is good. Like if I run a game with religious themes, I would probably disclose that I'm agnostic and thus my interpretation of certain things might be biased one way or another. If I don't disclose it, it doesn't technically harm anyone, but I think a disclosure makes it easier to regulate potential conflicts. Maybe it'll prevent people to whom the topic is close to heart from joining and then being disappointed, maybe it'll help not to blindside someone when they discover my views.

Of course generative AI and religion are not the same, but I think the general principle applies - just like it would with, say, politics, or basically anything that would commonly cause heated debates. Even if to someone it might just be a part of life with no specific meaning, to someone else it is a matter of an ideological stance. So, I only see benefits in disclosing such things.

(I was trying to think of a better example than politics and religion, but I don't think there ever been a tool quite as widespread and controversial as this in the times of the web.)

My two cents, because I read "training" being mentioned in the context of NotebookLM.


NotebookLM is a Gemini tool that uses retrieval-augmented generation (RAG) to vectorise and make searchable the data you upload. It processes and indexes the source material and then supplies the context to Gemini (thus artificially bypassing the 1 million context window limit). Right now, according to their ToS, they don't train on users' uploads.

There are other similar tools; Notion has a similar function, TextCortex is another, and Nouswise comes to mind as well. And I'm pretty sure ChatGPT and Claude have something similar with their projects folders. However, if you seek to retain total control of your data, you could always create your own system by running one of the open-source models locally (if you can) or on a cloud instance.
Since these systems are very good at rule-based decision-making (hence RPGs), you can upload a manual to index it, and now you don't have to go around and search throughout hundreds of pages for a specific rule the editors forgot to add to the general index.
To avoid hallucinations, the system by default is forced to anchor the output to the uploaded sources, plus, they usually produce a link as a footnote to the relevant quoted material.

I already covered this, but, again, if/because ppl want to know, or asked.

I might have misunderstood one of your points, so let me clarify mine. I don't think there's any imperative to disclose in advance, assuming you have legitimate reason to believe the data won't be used in training. However, if asked, I believe you shouldn't hide the use. In that situation, I suspect we're aligned.

I was thinking only of the complying in advance. :)

Won't it be easier to disclose in advance though? If there is a good chance that someone would like to know, then doing it right away just saves the hassle of them having to ask or being blindsided when they discover it another way. Plus it'd ensure that the only players in the game are the ones who are alright with it (or failed to read the disclosure), so the GM won't have to risk someone quitting over it.

It is preference at the end of the day, but I don't see the cons of such a disclosure. Extra time spent on the recruitment post, maybe.

So if, for example, it became a community/forum rule that games had to disclose up front whether or not AI is used in any capacity for that game (in this hypothetical, that would be because enough ppl on the forum care and requested such a community rule), you would disclose it up front?

I don't want to see AI used in play-by-post just like I don't want to see it used elsewhere, but it is used more and more here on GP, which makes me sad. Completely aside from the genuine ethical, financial, and environmental drawbacks, there is another reason I will never use AI to write or illustrate stuff myself. I am a creative; LLMs do not do a better job than I do at realizing my ideas. The effort I put in to realize my ideas in writing and illustration is what is fun and rewarding. I enjoy the process, and I get better as I do it. But I know I am an outlier, feeling more extreme about it than many others; as a writer, the first thing I do with a new device is turn off every spellchecker, autocomplete, grammar suggestion feature, and so on. My successes and my mistakes are my own; I don't want to be coddled by software.

So I won't use AI, but I understand that others will, who can justify their choice with themselves. If you choose to use AI, I urge you to use it in moderation. Please. I can attest from my observations in the workplace that the cognitive decline in people using LLMs frequently/extensively is very, very real.

Expecting me to disclose in advance is passing the onus from the one who cares to the one who doesn't. If it is something you are uncomfortable with, it is your responsibility to verify. I hear what you're saying about "Isn't it easier," but, as we're talking about a victimless choice (using specific LLMs that do not use your inputs for training), my opinion is that there is no need to disclose in advance.
If that's the rule, that's the rule. If it isn't, it isn't. Like I said, I'm not about complying in advance.

Eh, fair enough. Although asking every single time is bound to become awkward. But that's what games with explicit "no AI" disclaimers are for, I suppose.

---

Let us both be outliers in a corner!

You should look into indie web initiative (things like Nekoweb or Neocities), most people on there are against genAI as well, and it's fun to see the creativity similar to that of early internet being recreated. At least it makes me feel less sad on the topic, so maybe it'll cheer you up too. :3

Out of curiosity: bookkeeping LLM users - did you know that your tool (NotebookLLM or any other) does not use user input for training or did you have to look it up?

Can't help but notice, you didn't answer my question.

Yes, I would. If the rule is to disclose, then I'd disclose. I wasn't trying to hide anything.

---

Spaceseeker, I agree with so much of that. I've actually brought your "Hand-drawn character art" to my live tables. I don't go to those extremes, but I do try to ensure everything I offer my players is my own

That said, I find NoteBookLM to be a godsend for those hefty OSR tomes to parse disparate information and find trends.