Meta admits to training LLM AI with terabytes of torrented copyrighted works.

[Quoth]

Originally Posted by salamanderjuice

It could be pirated lectures or lecture notes or a set of books. Doesn't really matter. Let's say expertise in web design. Is it stealing if the employee uses their "unlicensed" knowledge of web design to build the company webpage?

That's not what LLMs are doing. The businesses directly and deliberately scrape content to build them. They have admitted it, but want the rules changed.

You are muddying the water with irrelevant comparisons.

And companies have been prosecuted (well, successfully sued for using trade secrets obtained from former employees deliberately stealing the information and then being hired, though that also isn't the same as your example.

EDIT:
The LLM makers want special treatment, they want new laws especially to benefit them doing stuff that people and companies have been sued for. No need to make up hypothetical scenarios that are not even equivalent. They have all admitted it.

[Quoth]

and
https://www.theregister.com/2025/03/...ata_copyright/

[SleepyBob]

Since every country has its own copyright laws, it seems like one potential fallout of this could be that useful LLM models can only be created in countries that make it fair use.

I'm not sure it's a great spot to be in for LLM's to be in widespread use in China, North Korea, and Russia, but nowhere else.