Hacker News Reader

Comment by rvnx | Hacker News Reader

Why LLM companies that depended on Anna's archive end up so clean ? Looks like Anna's archive was doing the dirty work, and the LLM companies were reaping the profits (and ironically still do, as they hold the largest databases of pirated content in the world).

Is it because the law doesn't apply to you when you have 1B USD ?

View on HN

While that may be the case it’s hard to make this claim when: - Anthropic settled a similar case - Anna didn’t show up in court

Showing up is a trap for Anna - who doesn't have 5 billion dollars to settle.

Justice should not depend on whether the aggrieved appears in court. That's a structural weakness of US law.

is there a country where if you don't show up to court you don't lose by default?

Exactly, how can you credibly mount a defence if you're unwilling to appear? Your defence is that, yours.

You cant mount defence if you dont have tons of money. If you are not super rich and can stay anonymous, showing up to money combat is just dumb.

Supposing that to be true... Does justice depend on what every other country is doing?

The legal system is a joke which received the amount of engagement it deserves.

As well as their domestic and national security apperatus, actually. Their own citizens are being used to propagate brain-computer interfaces amongst their own people.

Uh, aren't you confirming his opinion with that? After all, Anna doesn't have the money to fight this in court

No. Anthropic fought and paid $1.5 billion in settlement and agreed to delete all the copyrighted material.

I'm confused here, how is this not even more of a confirmation?

Essentially: have funny amounts of money and the law ceases to matter. Or don't, and be squashed by the right holders

$1.5 billion is more than $19.5 million though.

Also if I were to guess the damages because of sci hub is higher than Anthropic training the models. I don't think I know anyone who didn't bought a book because the summary is available or they can ask about it to AI.

Delete? Wasn’t that material already used to train models?

All AI companies should be forced to re-train their models without the offending materials, and this should also extend to all LLMs distilled from models exposed to copyrighted works. Also cover code under licences such as GPL as well. Not to mention patents and designs. This whole LLM business is a giant IP laundromat.

One of the best things about it IMO -- or we'd be spending the next hundred years waiting for copyright reform

"Deleting" data they already ingested is meaningless.

well i guess its copyright not distill-statistical-model-from-it-rights.

Anthropic knows they could just pay off the aggrieved party.

The operators of Anna's know they will go to prison.

A couple of millions are enough to grant you special treatment. Especially when you know where to "invest" some of them.

You can make an argument that training an LLM on something is not the same as copying it in the same way that your brain is not in breach of copyright for having watched a Disney movie. I'm not sure of the rights and wrongs of that but it complicates legal action.

Can I download an archive of movies so a human animator can study the techniques there?

Surely you have to make the copy to feed it into the llm for training, so

I think some of the LLM companies have used legally purchased materials.

Do you happen to have an example? The closest I can think of is adobe for images, but I've never heard of a text based llm trained purely on legally acquired books.

There's stuff here on Anthropic https://www.reddit.com/r/books/comments/1lkv2r9/anthropic_de...

Distribution. Anna's archive actively distributes the pirated material. LLM companies don't.

I'd argue the LLMs certainly distribute copyright material. That's why it can do things like:

https://g.co/gemini/share/20843b4609d9

Now, you could argue quotes are fair use - but can you argue the material isn't part of the LLM?

Fruit of the poisonous tree.

Reply on news.ycombinator.com