Writing for LLMs

One of the objections to LLMs like ChatGPT is that they violate copyright. The underlying model is the result of scraping an enormous amount of text. Sometimes, some of that text spits back out, more or less verbatim. The more you tailor your prompt to generate apparent copyright violations, the more you see it.

Fine–if I were an industry reliant on milking my copyrights, watching a new technology threaten to undermine my entire business model, I’d probably sue, too.

Take the financial aspect out of it for a minute, though. It’s awfully appealing that there are chatbots out there, with millions of users, that rely–however extremely slightly–on your own writing.

One of the main reasons that people write is to have some kind of influence, broadly defined. Even if you’re not trying to win people over to some particular opinion, you probably want them to see some small part of the world more like the way you see it. There will always be limitations, though: Even if everyone on earth can find out what you have to say, most of them won’t. Of the tiny percentage who do, far fewer will pay close enough attention, or come across the right bit of text at the right time, for you to have any influence.

But, if you aren’t thinking about copyrights, freelance fees, and the personal respect gained from a byline, LLMs serve as enormous syndication machines. Even though we don’t know exactly why the algorithms work, or how they learn, it’s clear that they work out patterns in text (and other inputs) that they ingest. It’s like a giant word association game, where the associations are determined by a vote of everyone who ever put pen to paper.

The crawling isn’t over. You’ll never be able to dictate how ChatGPT responds, but you can vote. Toss your insights into the maw of the next giant web-scrapes. All you have to do is publish them.