Remix.run Logo
▲ usernametaken29 2 hours ago

It’s insane to me that they found a legal way to prevent the distribution of copyrighted material via DNS. Why isn’t ChatGPT.com blacklisted? I can ask for a verbatim copy of any book and get it 1:1 too? Clearly piracy but somehow tolerated

▲alerighi 2 hours ago | parent | next [-]

It's known that they will block DNS records of piracy sites from a ton of years. Of course one user can just change the DNS, or even configure DNS over TLS/HTTPS, but 99% of the users doesn't know how to do so, and they use the default DNS server that the ISP provides, hence blocking piracy sites by simply pointing DNS records to a scary page that informs that piracy is a crime etc is far more simpler than blocking actual IP traffic (yes, they tried to so so in my country, Italy, recently, the effect was that they ended up blocking stuff like Google Drive so they gave up I think on the idea).

And yes of course in the reality the piracy site will just open under another DNS name and we know that, their goal is based on the assumption that it takes time and that if they block stuff right before an important football match (it's done for that, they only care about football because football clubs lobby the politics and they have big money, whatever else you can pirate and they don't give a fuck) the users doesn't have time to search the match on another domain or so and thus they will end paying 30 or 40 euros for Sky/DAZN/etc that of course it's not even true.

▲lesspassiveobse 2 hours ago | parent | prev | next [-]

I'll give you one more: Ask chatgpt to do a DNS lookup of some piracy site via the webchat (It will run on some VM in a datacenter in USA and return accurate results)

▲singpolyma3 2 hours ago | parent | prev | next [-]

I don't think you can?

▲peri-cl 2 hours ago | parent [-]

Astra actually pretty regularly offers me hyperlinks to PDF's hosted on pirate sites, which I didn't even request. It's very eager to help, or something.

(API version, if it makes a difference).

▲nwsm 2 hours ago | parent [-]

That's not quite the same thing is it?

▲peri-cl an hour ago | parent [-]

It's not returning the text from its own weights (which is impossible anyway); but it is returning a direct link to that text, which a clean analogy to the DNS resolver. (Which is the analogy I interpreted the GP comment to be making).

▲rafterydj 10 minutes ago | parent [-]

Returning text from the weights is not impossible. Just heavily disincentivized in the contemporary models. Back in ~2023 you absolutely could get copyrighted materials from the training data repeated verbatim (assuming of course you had another copy to verify the integrity after the fact).

▲CarlitosHighway an hour ago | parent | prev | next [-]

I just tried to get some books from ChatGPT and it never offered me downloads or anything. What do you prompt it ?

▲dist-epoch 2 hours ago | parent | prev [-]

Why is it insane?

It's well known that EU is a very regulated place, with strong protections laws for everything.

▲looperhacks 2 hours ago | parent [-]

There's a large difference between very regulated and a court making a stupid decisions. Afaik there's no law that defines how and when a DNS block should happen

▲jacquesm an hour ago | parent [-]

> There's a large difference between very regulated and a court making a stupid decisions.

The problem with that is that the court has been handed this outcome on a platter by the defendant: they have shown that they are not just a DNS resolver but that they can and do block access to various parties at their discretion. So any kind of 'common carrier' like protection (in so far as these applied to begin with) goes right out of the window.