We talk about “the Nora Ephron problem” where you’ve trained an LLM on her work and ask it to write you a rom-com in her style. To me, that feels gross and wrong. But any human rom-com writer is also influenced by Ephron’s work. It’s part of the culture. - ThreadSky

johnaugust.bsky.social • 99 days ago

We talk about “the Nora Ephron problem” where you’ve trained an LLM on her work and ask it to write you a rom-com in her style. To me, that feels gross and wrong. But any human rom-com writer is also influenced by Ephron’s work. It’s part of the culture.

Comments

johnaugust.bsky.social•99 days ago

I don’t think material generated by LLMs should be substituted for human work, nor have its authorial protections (copyright and otherwise). That’s a fight worth fighting. But we can’t wish LLMs away. They’ve already read everything.

mikalatos.com•99 days ago

They'll go away after strip mining the environment I imagine

johnaugust.bsky.social•99 days ago

I don’t think they’re going away. For all their many many faults, they are genuinely useful in areas outside of creative work.

mikalatos.com•99 days ago

Oh I think for sure there are great uses for AI. I don't know that generative AI is worth the current price tag though

johnaugust.bsky.social•99 days ago

Fair. I just think it’s important to acknowledge that LLMs aren’t just about creating text and images. They’re also doing things that feel like analysis or Google searches.

mikalatos.com•99 days ago

Yeah, we're in agreement for sure. And I appreciate that you're asking the moral questions, which are largely being brushed aside.

buckleystuff425.bsky.social•99 days ago

Respectfully, I disagree. A human being reading or influenced by Ms. Ephron still emerges with work created by the prism and filter of their own experience. LLM has no experiences other than that it strips from the work of others.

johnaugust.bsky.social•99 days ago

I think we agree. A human writer has the experience of being alive, of being in love. But is training an LLM on Ephron’s work legally and/or morally wrong?

buckleystuff425.bsky.social•99 days ago

I think there is. In that case, an LLM is essentially an aggregator which is making "choices" solely based upon choices exclusively made by Ms. Ephron. Even training LLMs on works in the public domain. Those writers have had no say in whether their works and especially their style should be sourced

johnaugust.bsky.social•99 days ago

I agree. How do we handle this? For a deceased author, should an author’s estate decide? Should the WGA? For screenwriters like Ephron, copyright for much of their work is held by the studios, which further complicates.

buckleystuff425.bsky.social•99 days ago

It would be wonderful if there was a blanket decision from WGA, voted upon by living writers. With deceased writers, I would like the WGA to step in as decisions by estates might undermine any decision made by the union. Hopefully it will be viewed as protection, not a stifling. Complicated issue

johnaugust.bsky.social•99 days ago

Because of the WGA, US screenwriters retain publication rights to their scripts, which is a foothold. But the issue isn’t writers vs studios as much as writers vs tech companies, so it will require government intervention. Difficult, obvs.

buckleystuff425.bsky.social•99 days ago

It's a conversation that no doubt will be ongoing.

garyjackson.bsky.social•99 days ago

I think you’re correct that it’s tricky legally, but we already have a word for a computer or person who, in the absence of talent, copies the style of an iconic artist:

A hack.

And the consequences of being hack *in a community of artists* has always been (and should be?) punishingly negative.

johnaugust.bsky.social•99 days ago

I like that “hack” is about both the process and the product. Still, how do we want to set rules and standards here? What’s acceptable (i.e. not-hack)?

garyjackson.bsky.social•98 days ago

It’s difficult for sure. Being “hack” relies on a mix of artistic community (process) and audience shaming (product), but the AI screenwriting goal seems to be “skip the process, anyone can make the product.”

Which is shameproof, because there’s no artistic integrity to be called into question.

garyjackson.bsky.social•98 days ago

In other words, what do writers do with someone who doesn’t *care* if they’re a good writer because they’re not a writer they’re a prompt engineer simply trying to crank out the most profitable collection of letters and punctuation?

davidwrites1.bsky.social•99 days ago

The tech companies themselves already draw a legal distinction between a human reading their content and synthesizing it verus an algorithm scraping it - look at all the “prove you’re a human” captchas. You can read as many LinkedIn profiles as you like, but they block/sue any scrapers.

johnaugust.bsky.social•99 days ago

Absolutely. I wonder how these distinctions will hold up legally and practically. Where is the boundary between AI scraping and a low-wage human worker paid to click?

davidwrites1.bsky.social•99 days ago

I feel like the tech companies will have two different answers to that (admittedly tricky) question, depending on whose content is being hoovered up, theirs or other people’s.

P.S. Thank you for being such an informed and thoughtful advocate for writers, particularly on this issue!

n-j-a.bsky.social•99 days ago

Moreover, writers cultivate a drive for originality, informed by experience, that weighs others' influence upon their works, and uses that to steer a course between the accepted canon and novel contributions to it. LLMs don't incorporate this drive at all, and prompting it seems impossible.

brooksreynolds.bsky.social•98 days ago

LLMs aren't people, why afford them the right to train and copy off any work?

And when is ownership going to be factored in? If AI creates it, no one has copyright, right? If WB prompts the next great franchise into existence, can't a random indie studio make its own spinoff?

johnaugust.bsky.social•98 days ago

Copyright-ability of material generated by LLMs should and will be a factor. But to me most, most of the issues aren’t legal as much as cultural. What is an ethical use of this technology, if any? Who decides?

Comments

Posting Rules

Reply