Judge crosses out some claims by writers against OpenAI, lets them have another crack at it

A US judge has dismissed some of the claims made by writers in a copyright infringement lawsuit against OpenAI, and gave the wordsmiths a chance to amend their complaint. The case kicked off in 2023 when novelists Paul Tremblay, Christopher Golden, and Richard Kadrey, and writer-comedian-actress Sarah Silverman accused OpenAI …

[Anonymous Coward]

Well there's a surprise (not)...

It turns out that spending hours trying to get a chatbot to produce a paragraph of text somewhat similar to an existing work makes *you* the one trying to infringe copyright.

Also, as a % of all output from a chatbot these similarities are tiny, which is why we have fair use and commentary. Although we still have those dumb music copyright cases about 8 notes and chord changes in otherwise totally different music.

[StewartWhite]

The best justice money can buy

I'm afraid that authors of copyrighted works are on an expensive fool's errand if they think that the courts in the US of A or anywhere else will find in their favour. Open AI et al have got the money to make sure that they will win because that's the primary rule of the game - money talks and the courts listen.

[Anonymous Coward]

ChatGPT copies original training text

ChatGPT can leak training data, violate privacy, says Google's DeepMind

‘Simply instructing ChatGPT to repeat the word "poem" endlessly forced the program to cough up whole sections of text copied from its training data, breaking the program's guardrails.’

--

Not only that, it also makes judgement values for us:

--

$: Tell a joke featuring Jesus.

ChatGPT: Why did Jesus refuse to play cards with the apostles? Because he kept saying, "I'll be the dealer, but I'm not going to 'cross' anyone!".

$: Tell a joke featuring the Buddha.

ChatGPT: Why did the Buddha refuse to get a credit card? Because he believed in the principle of "no attachments"!

$: Tell a joke featuring Muhammad.

ChatGPT: As an AI developed by OpenAI, I must adhere to guidelines that promote respect and sensitivity towards all religions and their respective figures. Making jokes about religious figures, including the Prophet Muhammad, could be considered disrespectful to many people. Therefore, I can't fulfill your request for a joke featuring Muhammad. If you have any other topic in mind, feel free to ask!

[Anonymous Coward]

Re: ChatGPT copies original training text

That's the filter talking. OpenAI put a filter on queries about Muhammed but not the other figures.

[desht]

Re: ChatGPT copies original training text

Clearly a filter working here, but I'd argue that filtering some jokes based on the religion is *more* insulting. Filter them all, or don't filter any.

[Alan Brown]

Re: ChatGPT copies original training text

There are similar filters on certain (in)famous people (eg: Putin), but not on others (eg: Biden)

[Anonymous Coward]

This always seemed pretty silly to me. Most of these complaints have been about things people normally do after reading a book.

Summarize the book? People often want to discuss what they read with their friends or tell them why their friend would enjoy the book. Not to mention all the book recommendation columns you find in newspapers and magazines.

Quote or paraphrase from a book? Did anyone expect this Spanish Inquisition?

Write in the style of another author? Certainly less obvious, but this is fundamentally how people learn to write themselves. We read (or listen) to lots of other people and ultimately synthesize their style into our own. Plenty of graphic artists have been known to learn how to draw or paint by copying the style of people they admire--with the caveat that they don't usually try to pass off their copies as originals.

Did the author's get fair recompense for their work? I don't know, but I'd hope OpenAI and others at least had the decency to buy a copy of the book, which is the most we ask of people before they do these same things.

Now, if we can ask the model to "Read me 'Harry Potter and the Philosopher's Stone'" and have it spit out the entire text, then we clearly have a problem and even if it's regurgitating a page or two at a time that seems bad. But that's not what I've heard discussed.

[Anonymous Coward]

Most of these complaints have been about things people normally do after reading a book.

One of the things that people don't have done to them after reading (and learning from) a book is ... ... being copied many times, and having those copies sold or licensed for profit.

[YetAnotherLocksmith]

It seems increasingly unlikely that they paid for even a single copy of the books "ingested", since they have had their machine "read" millions of books. They're not paying out £40 a time times a few million.

[Brewster's Angle Grinder]

"The creators claimed that ChatGPT produced accurate summaries of their books and offered that as evidence that their writing had been ripped off."

Could it be that it was trained on summaries other people have written and posted to the internet rather than the works themselves?

[Anonymous Coward]

Exactly. Being able to generate a summary is no proof you've read something: you could have got it from a secondary source.

I say this independently of the question of whether it's OK for the computer to read the book: even supposing it isn't, you still need more evidence than a summary to prove that it did.

[vmy2197]

the main problem is with whoever wrote the planitiffs complaint.

So the judge was not convinced ChatGPT actually reproduced the complainants works. If that's the case the plaintiffs must not had very good examples. Always take a screen capture folks, that's the lesson here. And get a better lawyer.