Given it's not data they trained on, I'm not sure the principles behind punishing them further.
This article misses this crucial detail.
2. The pirated information wasn't even used! It was just stored in a library, but was not used to train any released models.
fxtentacle•4mo ago
That seems very reasonable. Otherwise, stealing money, lending it out for interest, and then returning the stolen money while keeping the interest would be a viable way to earn money.
"The cost of building a company on theft should be the company itself."
Also seems reasonable, because it is the same rule applied to AI companies.
I'd say leave 20% for the VCs and split the other 80% of Anthropic stock among the authors of affected books. That would surely discourage similar behavior in the future.
freejazz•4mo ago
Copyright infringement plaintiffs are entitled to disgorge any profits a defendant made by infringing. As I pointed out upthread, Anthropic did not train on this dataset and therefore did not profit from it. The authors were instead seeking statutory damages for the pirating of the library.