Comment by darawk | Hacker News Reader

Comment by darawk | Hacker News Reader

Parent Full thread

> Why would I ever let bing crawl my site if they aren't going to send any visitors to me?

I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn

I mean, they could be nice and respect your robots.txt, but they certainly don't have to.

> fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate.

It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use. If it were spitting out exact copies of things it had read, it would probably be pretty easy to train that behavior out of it.

> I do love these imaginary scenarios where ChatGPT is going to find me the best air fryer, though. Where is that information going to come from, exactly? Barely anyone is making money writing reviews today, it's mostly farmed content. What happens when even those sites' reviews are quickly scraped and put into the next model iteration? Bing is going to have to come up with some kind of radical revenue sharing too if they want anything fresh.

I do agree with this, though. The LLMification of search is going to squeeze revenue for content creators of all kinds to literally nothing, at least if that content isn't paywalled. Which probably means that that's exactly where we're headed.

> I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn

> I mean, they could be nice and respect your robots.txt, but they certainly don't have to.

That case was limited to the CFAA, but you seem to get the gist of what I'm saying when I specified it's different when it's Microsoft doing the scraping. If Bing starts ignoring robots.txt and data still start showing up in their results, all the early 2000s lawsuits are going to be opened back up.

> It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use.

Unless there's a reason for them to be considered fair use, derivative works are going to lose a copyright suit. And what's the fair use argument? If I'm the only one on the internet saying something and suddenly ChatGPT can talk about the same thing and I'm losing money as a result, there's no fair use argument there. Search engines won those early lawsuits by being transformative (index vs content), minimal, and linking to their source. None of that would apply here.

What GP means is that ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative. Instead, it resembles what you'd get if you asked a (credulous and slightly dumb) human to read a selection of documents and then summarize them. These kinds of summaries are absolutely not copyright violations, even if the source document can actually be identified.

> ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative.

Isn't this exactly what a court case would be trying to clarify? If so wouldn't assuming this be begging the question?

There exist other laws, jurisprudence, and even entirely different judicial systems besides those currently used in the USA!

Sadly, seem like the decision in that case was changed. From your link:

> In a November 2022 ruling the Ninth Circuit ruled that hiQ had breached LinkedIn's User Agreement and a settlement agreement was reached between the two parties.

It wasn't changed, it's just that there's more than one issue at hand: the earlier decision was that hiQ didn't violate CFAA, the later one was that it did violate LinkedIn's EULA. The November 2022 ruling specifically states that hiQ "accepted LinkedIn’s User Agreement in running advertising and signing up for LinkedIn subscriptions" - keep in mind that LinkedIn profiles haven't been public for a while in a sense that logging in is required to view them, and thus to scrape them.

Hence why OP is saying that this all will lead to increase in paywalls and such, and a reduction in truly public content.

Reply on news.ycombinator.com