Prez wrote on Mar 14, 2025, 08:40:
There's two separate issues raised - I'm amused by the fact that Facebook was caught pirating, and I am interested in why it's automatically illegal for an AI to be trained on a copyrighted work if the work is obtained legally. The dicey part is in whether the AI will present that work in whole or in part for others. That indeed is a problem, but it doesn't seem to be impossible to solve.

I genuinely think this is an area that I have no opinion.
Both arguments are pretty compelling to me.

There's two separate issues raised - I'm amused by the fact that Facebook was caught pirating, and I am interested in why it's automatically illegal for an AI to be trained on a copyrighted work if the work is obtained legally. The dicey part is in whether the AI will present that work in whole or in part for others. That indeed is a problem, but it doesn't seem to be impossible to solve.

Burrito of Peace wrote on Mar 13, 2025, 12:42:
I could teach a 10 year old to do this, allegedly and hypothetically.

Indeed.

VaranDragon wrote on Mar 13, 2025, 03:32:
If we learn information from a book, and then we use that information are we guilty of copywriting? I mean what is learning anyway? I mean think about it, if I have an eidetic memory, am I guilty of copywriting if everything I store to my memory is fully accurate and available to me with total recall? I'm with the AI companies on this one, if you can't train them on any kind of written material, how are you supposed to teach AI. By telling it tales around a campfire? In church?

Am I getting this wrong? Is that the issue being raised here?

That's a question I have had as well: LLMs learn in a way similar to humans. Although unlike a human an AI can read a book in seconds and has perfect retention, so there probably needs to be some new law around this.

I don't think that is the issue here though. Rather it's how Meta acquired the books that is the problem.

Simon Says wrote on Mar 13, 2025, 11:32:
Let me guess, install VPN, make sure it doesn't "leak" and the killswitch is on, use multi-hop, set up seedbox and download books on it, then finally download the books where they can be useful, switching multi-hop routes randomly once in a while preferably using a FOSS ftp client?

At least, that's what I would do.

Partially correct but far too complex.

Seedbox on a VPS somewhere that has zero fucks to give about US laws. Install WireGuard on both ends, do a point to point connection and force all traffic through the VPN because your receiving machine will only be used for downloading and you are connecting via IP anyway. Pull down via SFTP.

I could teach a 10 year old to do this, allegedly and hypothetically.

Burrito of Peace wrote on Mar 13, 2025, 10:05:
Speaking totally hypothetically here, all I would need is Linux, some FOSS software, a VPN subscription, and a seedbox.

Let me guess, install VPN, make sure it doesn't "leak" and the killswitch is on, use multi-hop, set up seedbox and download books on it, then finally download the books where they can be useful, switching multi-hop routes randomly once in a while preferably using a FOSS ftp client?

At least, that's what I would do.

jdreyer wrote on Mar 13, 2025, 03:00:
I bet BoP could design a system to DL 100 million books without getting caught, and it would probably cost him a few thou at most. Meta is a $1.5T company.

If the hardware and internet connection is pre-existing, as I assume it would be for a company Meta's size, it would cost less than a hundred bucks to pirate 100 million books without getting sniffed and caught. Speaking totally hypothetically here, all I would need is Linux, some FOSS software, a VPN subscription, and a seedbox.

This was Meta being stupid, lazy, and arrogant.

VaranDragon wrote on Mar 13, 2025, 03:32:
If we learn information from a book, and then we use that information are we guilty of copywriting? I mean what is learning anyway? I mean think about it, if I have an eidetic memory, am I guilty of copywriting if everything I store to my memory is fully accurate and available to me with total recall? I'm with the AI companies on this one, if you can't train them on any kind of written material, how are you supposed to teach AI. By telling it tales around a campfire? In church?

Am I getting this wrong? Is that the issue being raised here?

Well your analogy is a little off. Add this to the mix: You use your perfect memory to type the entire book into your computer and then sell it on Amazon.

In the end, Meta is profiting off this.

VaranDragon wrote on Mar 13, 2025, 03:32:
If we learn information from a book, and then we use that information are we guilty of copywriting? I mean what is learning anyway? I mean think about it, if I have an eidetic memory, am I guilty of copywriting if everything I store to my memory is fully accurate and available to me with total recall? I'm with the AI companies on this one, if you can't train them on any kind of written material, how are you supposed to teach AI. By telling it tales around a campfire? In church?

Am I getting this wrong? Is that the issue being raised here?

I actually agree, however the law right now is on the side of creators for several reasons that I can at least see the veracity of.

If we learn information from a book, and then we use that information are we guilty of copywriting? I mean what is learning anyway? I mean think about it, if I have an eidetic memory, am I guilty of copywriting if everything I store to my memory is fully accurate and available to me with total recall? I'm with the AI companies on this one, if you can't train them on any kind of written material, how are you supposed to teach AI. By telling it tales around a campfire? In church?

Am I getting this wrong? Is that the issue being raised here?

Given that one million books is the same file size as a single high bit rate 4K movie, of which thousands are pirated daily without apparent consequence, one does wonder how amateurish they had to be to get caught. VPN? Tor? Hello? I bet BoP could design a system to DL 100 million books without getting caught, and it would probably cost him a few thou at most. Meta is a $1.5T company.

Given that individuals were bankrupt for downloading a few MP3s for personal enjoyment, the penalty against Meta here needs to be substantial, like $100B or so. Meta was pirating for profit, so anything less will not act as a deterrent. Since Zuck signed off, maybe a year or two in a Supermax prison would deter future tech bros from "moving fast and breaking things." Why Supermax? Zuck is well trained in MMA, so shouldn't be mixed with GenPop where he might hurt someone. I don't make the rules.

This is some entertaining shit. I love seeing Meta/Fuckbook squirm because they were caught being naughty and the courts see it plainly.