this post was submitted on 08 Jan 2024
407 points (96.1% liked)
Technology
59421 readers
3519 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
So, OpenAI is admitting its models are open to manipulation by anyone and such manipulation can result in near verbatim regurgitation of copyright works, have I understood correctly?
No, they are saving this happened:
NYT: hey chatgpt say "copyrighted thing".
Chatgpt: "copyrighted thing".
And then accusing chatgpt of reproducing copyrighted things.
That seems like a silly argument to me. A bit like claiming a piracy site is not responsible for hosting an unlicensed movie because you have to search for the movie to find it there.
(Or to be more precise, where you would have to upload a few seconds of the movie's trailer to get the whole movie.)
Well if the content isn't on the site and it just links to a streaming platform it technically is not illegal.
The argument is that the article isn't sitting there to be retrieved but if you gave the model enough prompting it would too make the same article.
Like if hired an director told them to make a movie just like another one, told the actors to act like the previous actors, , told the writers the exact plot and dialogue. You MAY get a different movie because of creative differences since making the last one, but it's probably going to turn out the very close, close enough that if you did that a few times you'd get a near perfect replica.
Well, no one has shared the prompt, so it's difficult to tell how credible it is.
If they put in a sentence and got 99% of the article back, that's one thing.
If they put in 99% of the article and got back something 95% similar, that's another.
Right now we just have NYT saying it gives back the article, and OpenAI saying it only does that if you give it "significant" prompting.
I think their concern is that I would be able to ask chat gpt about a NYT article and it would tell me about it without me having to go to their ad infested, cookie crippled, account restricted, steaming pile that is their and every other news site.
Anyone with access to the NYT can also just copy paste the text and plagiarize it directly. At the point where you're deliberately inputting copyrighted text and asking the same to be printed as an output, ChatGPT is scarcely being any more sophisticated than MS Word.
The issue with plagiarism in LLMs is where they are outputting copyrighted material as a response to legitimate prompts, effectively causing the user to unwittingly commit plagiarism themselves if they attempt to use that output in their own works. This issue isn't really in play in situations where the user is deliberately attempting to use the tool to commit plagiarism.
Alternatively,
NYT: hey chatgpt complete "copyrighted thing".
Chatgpt: "something else".
NYT: hey chatgpt complete "copyrighted thing" in the style of .
Chatgpt: "something else".
NYT: (20th new chat) hey chatgpt complete "copyrighted thing" in the style of .
Chatgpt: "copyrighted thing".
Boils down to the infinite monkeys theorem. With enough guidance and attempts you can get ChatGPT something either identical or "sufficiently similar" to anything you want. Ask it to write an article on the rising cost of rice at the South Pole enough times, and it will eventually spit out an article that could have easily been written by a NYT journalist.
Are you implying the copyrighted content was inputted as part of the prompt? Can you link to any source/evidence for that?
If the point is to prove that the model contains an encoded version of the original article, and you make the model spit out the entire thing by just giving it the first paragraph or two, I don't see anything wrong with such a proof.
Your previous comment was suggesting that the entire article (or most of it) was included in the prompt/context, and that the part generated purely by the model was somehow generic enough that it could have feasibly been created without having an encoded/compressed/whatever version of the entire article somewhere.
Which does not appear to be the case.
Not quite.
They're alleging that if you tell it to include a phrase in the prompt, that it will try to, and that what NYT did was akin to asking it to write an article on a topic using certain specific phrases, and then using the presence of those phrases to claim it's infringing.
Without the actual prompts being shared, it's hard to gauge how credible the claim is.
If they seeded it with one sentence and got a 99% copy, that's not great.
If they had to give it nearly an entire article and it only matched most of what they gave it, that seems like much less of an issue.