Hacker News Reader

Comment by nicce | Hacker News Reader

I guess they must delete all models since they acquired the source illegally and benefitted from it, right? Otherwise it just encourages others to keep going and pay the fines later.

View on HN

In a prior ruling, the court stated that Anthropic didn't train on the books subject to this settlement. The record is that Anthropic scanned physical books and used those for training. The pirated books were being held in a general purpose library and were not, according to the record, used in training.

That is something which is extremely difficult to prove from either side.

It is 500,000 books in total so did they really scan all those books instead of using the pirated versions? Even when they did not have much money in the early phases of the model race?

The 500,000 number is the number of books that are part of the settlement. If they downloaded all of Libgen and the other sources it was more like >7Million. But it is a lot of work to determine which books can legitimately be part of the lawsuit. For example, if any of the books in the download weren't copyright (think self published) or not protected under US copyright law (maybe a book only published in Venezula) or it isn't clear who own the copyright then that copyright owner cannot be part of the class. So it seems like the 500,000 number is basically the smaller number of books for which the lawyers for the plaintiff felt they could most easily prove standing.

So how did they profit off the pirated books?

According to the judge, they didn't. The judge said they stored those books in a general purpose library for future use just in case they decided to use them later. It appears the judge took much issue with the downloading of "pirated content." And Anthropic decided to settle rather than let it all play out more.

But how the settlement cost was then defined if nobody read those books and there was no financial lost...

Reply on news.ycombinator.com