Datasets:

---

nyuuzyou

/

archiveofourown

like 14

Oh, rad.

As someone who's "from the other side," I do think licensing it BY-NC-SA (specifically non-commercial share-alike) would go some way towards goodwill with the community. There's a conception (founded or not) that the purpose of using AO3 data in an AI context is to monetize it; that's where a lot of the ire is coming from. Folks feel gross that their hard work, which they're sharing for free (and which there is great community mythos about never monetizing ever), might be incorporated in a model to make something for profit.

You will ofc get the copyright hardliners and "replacement" fearmongering still, but I see the potential for constructive dialogue with at least some people.

Oh, rad.

As someone who's "from the other side," I do think licensing it BY-NC-SA (specifically non-commercial share-alike) would go some way towards goodwill with the community. There's a conception (founded or not) that the purpose of using AO3 data in an AI context is to monetize it; that's where a lot of the ire is coming from. Folks feel gross that their hard work, which they're sharing for free (and which there is great community mythos about never monetizing ever), might be incorporated in a model to make something for profit.

You will ofc get the copyright hardliners and "replacement" fearmongering still, but I see the potential for constructive dialogue with at least some people.

Their work is incorporated in a model to make something for profit. Unfortunately, the guys doing it don't publish datasets, they just scrape and bake. And they don't care about licenses, so...

I'll put it like this. If organizations train on the data without a license - as in, the least possible permissions - granting permissions isn't going to help, is it? A license can't give less permissions than no license at all.

I am not going to get into the question of what's right or wrong, but I feel something really needs to be noted here. So people understand the situation.

The fact is, billions of dollars have been invested into just the training of SOTA models by OpenAI, Google, Meta, Deepseek etc, and these models were trained on a vast corpus of copyrighted text. Any legal argument that someone cannot train a model on data without explicit permission - permission they would not have by default - is tantamount to arguing that every major LLM in existence should be illegal, and all those companies billions of dollars of training time should be considered burned in a trash can. Call me cynical, but... that's an uphill legal battle, I'll put it like that.

And that's before we consider the literal strategic implications, because even if people somehow won such a court case, it would only apply in some (or one) countries. Leaving others to continue advancing LLMs while crippling whatever nations supported the argument. If China made it illegal for some reason, Deepseek, Qwen, etc is crippled, while America advances. Or if America makes it illegal, Deepseek and Qwen advance unfettered while we're set back. That'd come up in that hypothetical court case too, no doubt.

Which means sure, you can takedown datasets. But the dataset exists because someone automated loading webpages. The people making money off AIs, definitionally, can afford to pay someone to whip up a scraper. And they do that, to the point that many sites complain its chewing all their bandwidth up. The only people this really inconveniences are hobbyists and enthusiastics sitting at home.

I do sympathize. I get why people are upset and I don't dispute that they have cause to want control. But this entire... thing... is fundamentally misleading people if it doesn't make it expressly clear that people "monetizing AI" are not going to be affected in any significant way whatsoever even if you take down every public fanfic dataset -- because there will still be a public dataset, known as the ao3 website, and they'll - financially motivated as they are - scrape it. That's the truth.

Their work is incorporated in a model to make something for profit. Unfortunately, the guys doing it don't publish datasets, they just scrape and bake. And they don't care about licenses, so...

Ugh, rate limit. You seem to think I'm somehow disagreeing with you. I'm not; you may have seen that I'm a fanfiction writer and strawmanned me in with the rest of them (whatever, everyone else is too). I'm actually an RMS-style free software weirdo and a staunch socialist... that is, I'm an even more annoying genre of guy

I'm discussing, as researchers (taking your word in good faith), your goodwill with the community - yours, not deepseek or openAI or meta or roko's basilisk or w/e. People have real emotional attachments to the work you're doing operations on; it would make the world a better place to at least consider their goodwill, even if they're being mean to you and don't understand that you're less of a dick than the others. Do you haaaaave to? No, it's the internet and anyone can functionally get away with anything. Does it create a better atmosphere under which to do your work? It does.

Some of the the behavior in here is unhinged and patently dehumanizing... see all those deleted posts in the pinned thread, calling fanfiction writers slurs, autists like the whole AI/ML scene isn't also crammed full of neurodiversity lmao, and saying "your sonic mpreg vore has negative value, but these people are doing REAL RESEARCH, f****t." There's folks who are openly gleeful at trampling over the boundaries of people whose work they consider lesser. One might be forgiven for lumping this community in with the meta assclowns based only on behavior. I'm taking this community at its word that it is not like them. (I even asked for use cases because I think cool things could be done here! Jesus.) I'd like it to show the same kindness the other way.

Their work is incorporated in a model to make something for profit. Unfortunately, the guys doing it don't publish datasets, they just scrape and bake. And they don't care about licenses, so...

Ugh, rate limit. You seem to think I'm somehow disagreeing with you. I'm not; you may have seen that I'm a fanfiction writer and strawmanned me in with the rest of them (whatever, everyone else is too). I'm actually an RMS-style free software weirdo and a staunch socialist... that is, I'm an even more annoying genre of guy [...] I'd like it to show the same kindness the other way.

I'm not strawmanning you and I hadn't seen that you're a fanfiction writer - I hadn't looked at all. And in fact, I have written fanfiction myself, albeit not on AO3.

My point was that it would be a deception to put a license on the dataset that implies people can't use it to train a certain kind of model, because licenses meant to restrict training are unlikely to be enforceable in any realistic world. Ergo, it is not a kindness. It misleads people into thinking a strategy is viable that isn't. And not just for the writers, but for people like nyuuzyou - I don't believe him putting a license he didn't get permission to apply is going to materially reduce the umbrage of people spamming DMCAs. Some would likely even find such a response more offensive, because it would still imply he is deciding, not them. The fanfic authors are not a monolith and "many of them probably think X" wouldn't avoid the fundamental conflict of who has the right to call the shot.

Of course, that's my subjective opinion, and I can understand (if not agree) if you think it 'communicating goodwill' outweighs that or that I'm wrong altogether.

Agree to disagree - I see your point especially re: a license all did not consent to, I just think throwing your hands up and saying "welp, we are already in a post-respect society" is a damn shame without trying to bridge that gap somehow. These are profoundly hurt people who think that this community, specifically, is making profit off of hundreds of thousands of hours of their collective work, which they rendered with love and without payment because of the idea that "if we ask for money, our community will be destroyed." There will always be the people who insist they have unassailable copyright over a work that uses large fractions of someone else's IP, there will always be the comp sci undergrads who think artists are dumb and deserve to be exploited, and there will be corporations making their CEOs rich for as long as society is willing to tolerate them. I think that we can all do better.