Hacker News Reader

Comment by briandw | Hacker News Reader

I'm really confused by the FSF statement here. The court ruled that the use of copyrighted information is fair use. The issue is that Anthropic pirated (obtained illegally) copyrighted work and that was the offense. FSF books are free to download and store etc. The license says: "This is a free license allowing use of the work for any purpose without payment." So how can they claim that their rights were infringed when the court ruled that the problem was the illegal downloading of copyrighted work? It's impossible to illegally download a FSF book.

View on HN

1. There are some uses of the copies that were not granted summary judgement

> And, as for any copies made from central library copies but not used for training, this order does not grant summary judgment for Anthropic. On this record in this posture, the central library copies were retained even when no longer serving as sources for training copies, “hundreds of engineers” could access them to make copies for other uses, and engineers did make other copies.

Whether or not those other actions met the requirements of the FDL is untested and would be the subject of a trial, had this gone to trial, but it didn't.

2. The FDL does have requirements that must be met for the use of copies to be permissible -- it doesn't allow you to do anything you want.

I interpreted this to be a class action to which they were a party, not something they principally launched themselves?

It is. I'm just commenting on why it isn't that straightforward that the FSF presumably wouldn't care. Copyleft is an exercise of copyright. The FSF doesn't believe in permissive use of works - they believe in using copyright licensing to force others to share the way they believe others should share.

While there are requirements on making copies in the FDL, I think it is extraordinarily unlikely that a court would find that a company making internal copies would violate the license when those restrictions are just along the lines of "and you must include a copy of the license".

And the FSF would be extremely foolish to ever pursue such a suit, because extremely ordinary non-AI related activities involving working with internal local documents also make copies in a similar way. If OpenAI violated the FDL by doing so then the FDL is a foot gun of a license that companies would be well advised to avoid.

The only suit that makes any sense would be the one against using the FDL licensed documents to train the not-FDL licensed AI... and the judge already rejected that in this case.

The requirements being easy to meet doesn't absolve someone from having to follow them. And the FSF clearly says they aren't pursuing a suit in this case. I'm not suggesting that filing suit here would make sense, rather, that the FSF does care deeply and fundamentally about requiring requiring copyleft restrictions -- they vehemently don't tolerate permissive use under their licenses.

> If OpenAI violated the FDL by doing so then the FDL is a foot gun of a license that companies would be well advised to avoid.

That has been said about a lot of FSFs licenses, and in fact, many companies do avoid them.

> The requirements being easy to meet doesn't absolve someone from having to follow them.

Everybody ignores some GPL requirements. For example the following one:

> a) You must cause the modified files to carry prominent notices > stating that you changed the files and the date of any change.

Look at any GPL project. Do they have "prominent notices" in all files modified by someone who is not the initial author? For every person who modified them?

How long would be the list of dates alone for files that are often modified?

Another great example of a footgun. Those terms did make more sense in the days before source control when it was common practice to do that anyway, and in many older projects you will find that they were followed.

Ultimately, the terms of the license do matter if someone wants to enforce them.

They said that the LLM can hold the information but can't produce it ipsis litteris. But if it produces a plagiarized version then it needs be settled with those who publish it.

That said, one might say it was unintentional and it would be impossible to verify all llm work with that premise.

I also found the statement bizarre. They don’t seem to have any argument for compensation of any kind unless the books were under a restrictive license that required derived works to also be open source.

No, even under such a restrictive license (which I think the GFDL is?) there's no argument.

Their copyright was not violated by anthropic downloading the books, because anthropic had a license to do that.

And their copyright was not violated by anthropic training on the books, because the court found that no ones copyright was violated by doing this. Antrhopic didn't need a license to do this. So the restrictive terms of the license can't prevent it.

I mean they might have an argument for compensation based on "well the settlement Anthropic agreed to didn't exclude us even though they didn't violate our copyright"... but just for the compensation outlined in the settlement.

I think they are implying that they believe models trained on their copyrighted information should be open-source.

Not 'open source' but 'free', of which RMS has very strong feelings about the difference.

Didn't understand it well, I thought it was because english is not my native language.

Reply on news.ycombinator.com