No, you are the one taking one specific implementation of AI and trying to generalize that to all AI. You are the one obsessed with the specific use of AI as a TV script writer, which I honestly could not care less about one way or the other. I’m interested in generative AI from the broader perspective. And I think the use of publicly-viewable information as data for training AI should rightly be considered fair use/transformative use. Restricting what data can be used in AI training will result in much weaker AI models if not outright kill powerful general-purpose AI. I realize that you think that is a positive outcome, and that AI is an evil plot from Capitalist Techbros, but I share neither that perspective nor your constant boiling rage.
…the claim that “AI doesn’t plagiarize” is a claim that has been shown to be untrue. I have no way of knowing if its limited to this one specific implementation, and neither does anyone else. It certainly won’t remain that way. Not if you want AI to “get better.”
Restricting the datasets to intellectual property that they have permission to use should be a no-brainer. It should have been a consideration from the outset. That is wasn’t, and that has left copyright holders scrambling to defend their work, is a the problem here.
Hubris and arrogance will kill “powerful general purpose-AI”, not people defending their intellectual property rights.
Anything that can be looked at by anybody on the internet can be looked at by anybody on the internet should be a no-brainer.
…“language learning models” aren’t “anybody.” They aren’t people. Its software. And that software is being used to promote the commercial interests of already rich people to the detriment of content creators.
Its telling that you admit if we removed everything that it didn’t have permission to use, AI would cease to become useful. That means the dataset has value. The millions of images and words that AI has “learned from” is valuable. On its own? Its worthless.
It’s the content that is important. Not the technology.
And lest we stray too far off topic: I doubt that this is what the Church is worried about here.
Yes. It’s an important social question in general, but it’s extremely unlikely that it’s relevant to @Velocity 's situation.
That isn’t something that I grudgingly admit to, it is something I state outright: the two things that make powerful general generative AI possible when it wasn’t in the past is the vast amount of computing power available and the vast amount of training data available. That doesn’t mean that the hundreds of millions of people whose collective internet posts make up those datasets should be paid for them, though, because it is only valuable as a collective.
…will be my last post here because this is getting off topic.
But it isn’t valuable as “part of a collective.” Its valuable because the human brain is amazing and unique and super-powerful and creates incredible things that a computer will never be able to do. Without us: AI is nothing. Worthless.
I wonder if any ToS’s are being violated by the way they crawl and retrieve data.
Not only must the user know that the Terms of Service exist, but they also must have actual or constructive notice that use of the website is subject to the Terms of Service. If not, the court can rule that the user was not aware that they were agreeing to terms, thereby making the contract invalid.
Proving your user accepted your Terms of ServiceAccording to contract law, in order for a contract to be valid, there must be an offer, consideration, and acceptance. Being able to prove your customer accepted your terms (and therefore entered a contract with you) will determine whether or not your Terms and Conditions are enforceable.
If your customer did not (or did not need to) accept your terms before using your site, purchasing your product, or using your software, they are not bound by your Terms and Conditions. If they did not accept, then your business will not be able to enforce any of the clauses within your agreement, including, for example, your clause to compel arbitration.
https://ironcladapp.com/journal/contracts/terms-and-conditions-legally-binding/
If the data is readable without agreeing to or knowing about the ToS, then the ToS isn’t worth much. (If the data was scraped from deliberately circumvented paywalls, there is probably a case, but the data being used by the AI training comes from Common Crawl, which has been around since 2007 and is a registered non-profit. If they were doing anything illegal I think they would have been in trouble before now.)
In response to the OP’s question, I’d say, “I leveraged my knowledge of current technologies to work efficiently. I’m not revealing my work methods, as I am a free-lance contractor with no requirement to teach this method to others.”
StG
His answer was that no one knows, and it is a favorite topic over lunch. And it depends a bit on the ChatGPT license, which might indemnify them from infractions by users.
But you guessed right about him not understanding AI. He is under the impression that the person who writes an AI program effectively writes all the things it produces. Which we both know isn’t close to true. My impression is that he is subject to the computer is just a big adding machine fallacy.
It’s a common problem with new tech - sometimes the law just isn’t caught up, and the people with expertise in law don’t understand the tech, and vice versa. With AI it’s an even bigger problem because we don’t really understand how they do what they do - not even the people who built them.
Regulating AI will be fraught with error, I’m afraid.
I give him a pass because the talk wasn’t about that. But I doubt the regulators and the legislators will do any better. And even if they do, some judge will screw it up.
CNN reports about that case too. I think it is unbelievable that a lawyer trusts ChapGPT blindly to prepare a case and I can only hope that the OP has checked what he gave his clients. It woud be very embarrasing if Chatty had made up some godly quotes.
On the other hand, it might start a new religion. I fear that it will not be better than the ones we already have, sadly.
Any update on how this worked out?