Much ado about Knotting: Omegaverse vs. AI

Oh boy. The internet is in uproar about omegaverse once again. Previously, we saw the omegaverse in a lawsuit that got into some weird stuff about copyright. (The suit was ultimately thrown out of court.)

This time, omegaverse has cropped up in a new legal context: AI.

Wait, What? Omegaverse in MY AI?

The best way to understand modern “artificial intelligence” is as predictive tools. Current tools like Chat-GPT are known as large language models, or LLMs.

What that means is these are tools that generate text, based upon prompts, by predicting the next most likely words. It’s not really intelligent, in the sense that it’s not thinking or creating or inventing. AI can’t reason. It can’t really do anything other than … predict what the next most likely text is (or hallucinate, but that’s another story).

So, what does this have to do with omegaverse?

Well, for LLMs to be any good at prediction, they have to learn. And to learn, they need to ingest huge databases called training sets. These sets are enormous, allowing the tool to “learn” how to best predict the next word or sentence or paragraph.

Where do these training sets come from?

Well, much of the training base for LLMs was scraped from the internet.

Enter Sudowrite

People had long speculate that LLM training bases included fanfiction. Fanfiction was prolific on the early internet, in part because it was a text-based internet, and partially because a lot of people who were online were nerds.

Fanfic also operates in a weird sphere where you can’t make money off of it, so it is freely available and distributed. Most fanfic sites are free to access. Stories aren’t behind a paywall or anything like that.

Some archives, like FF.net, have existed since the turn of the millennium. AO3, which has been around since the mid-aughts, hosts hundreds of thousands of fics.

That’s a pretty tempting resource when you need trillions of words to input into your training base.

A new writing tool called Sudowrite seemingly proved that fanfiction was part of its dataset, as it seemed to start writing omegaverse fanfiction when given certain prompts. (People have noted that certain LLMs are … strangely horny, with almost everything devolving into boning.)

What’s Knotting Got to Do with It?

If you remember the omegaverse copyright debacle, one of the things the press fixated on was knotting. Knotting in omegaverse is an “alpha” trait, where the alpha’s cock swells up at the base. This is used to “knot” the omega, or tie the two partners together.

The idea of the knot is based on dogs and wolves. This is an evolutionary mechanism to try and ensure paternity, and fanfic writers adopted it into omegaverse — aka “wolf kink porn.” (It’s not really wolf kink porn, but for the uninitiated, it sure can read like it—and there are a fair few werewolf or “wolf shifter” stories that use the trope.)

Omegaverse started out in fandom. Writers have since adopted it for original fiction. But that means that there’s a good chunk of omegaverse fanfiction out there, freely available on the web.

Plagiarism and Copyright Issues Don’t End There

Of course, LLMs have other issues with copyright and plagiarism. Sudowrite, for example, seems to be able to generate the first few pages of Dan Brown’s bestselling novel The Da Vinci Code. Of course, that could mean Brown’s just an extremely predictable writer (a valid assumption).

Or it might mean that someone fed The Da Vinci Code into Sudowrite’s database. This isn’t a totally unfounded accusation. At one point, Sudowrite literally encouraged writers to input their entire manuscript into the machine in order to get a “blurb” they could use for marketing.

Those works are under copyright, whether they’ve been registered with the US copyright office, whether they’ve been published or not. Copyright is inherent, so if Sudowrite added those works to their training database, everything generated with this tool is committing copyright infringement.

More blatant evidence of such behavior comes from Chat-GPT, which was fed novels from SmashWords. Seemingly, someone created a database that included “free” books, starting in 2014. Of these, thousanda are clearly under copyright.

Nonetheless, this material was all added to Chat-GPT’s training database. Once again, this is all copyrighted material—and that means everything you generate with GPT is committing copyright infringement.

This is probably the biggest reason why it would be immoral to allow works generated with AI to be copyrighted—which is a whole other kettle of fish. (If you can’t copyright what you generate, it’s fair game for someone like me to take it and put it up for sale elsewhere, to generate money for me.)

What about Fanfic?

Fanfic exists in a weird copyright space, as I alluded to earlier. Since it’s a derivative of copyrighted work, you can technically be sued over it.

This used to be quite common, actually! It’s why you’ll see older fanfics adorned with cutesy “disclaimers” like “don’t own, don’t sue.” Usually, the writer would disavow that they were making money off the fic, which then puts it into a gray area in terms of law.

See, copyright infringement happens if you make money off something too similar to another product. If you write Mickey Mouse fanfic and sell it, Disney can sue over copyright infringement. They can say that your product is confusing the market and cutting into their profits—that is, people who buy your fanfic may not realize they’re not getting “genuine” Mickey Mouse.

This is why there has been a long-held standard against fanfic writers and fan artists making money off their work.

In some cases, copyright holders didn’t want you to do fanfic, even if you weren’t making money off it. It’s only been in the last fifteen years or so that fans have gained enough power to drop disclaimers; I rarely see them now.

That said, copyright holders could most definitely decide to sue again.

And they might, especially since fanfic is now being absorbed into generative AI tools like Sudowrite and GPT. The copyright situation with those generations is murky, as I said, but already people are dumping AI-generated novels into Amazon’s self-pub program in a bid to make money.

Omegaverse to the Rescue?

The discovery of omegaverse tropes in Sudowrite’s training database lends more strength to legal challenges to AI. Since fanfic in particular is in such a murky legal situation, it follows that anything trained on fanfic is similarly shady.

The discovery of the Smashwords novels is even more damning, of course, but the fanfic situation presents a very challenging legal problem. If fanfic is technically a violation of copyright we all simply tolerate because it’s free, then we can’t possibly charge for AI generated text where fanfic was included in the training base.

More than that, the use of fanfic moves us into the realm of moral rights. Even if fanfic is freely available, that doesn’t mean you can just up and do whatever you want with it. Authors and artists have certain moral rights to their work, which allow them to control how the work is used. This prevents people from, say, using someone’s image to promote fascist groups.

So, even if fanfic authors can’t lean on copyright law, moral rights arguments still protect those works. They also protect copyrighted works, of course.

So, three cheers for the omegaverse, for laying bare what we all suspected. Generative AI tools are nothing but hotbeds of plagiarism and copyright infringement. The lawsuits are going to be fun.