2) The models explicitly don't have every book inside.
At best (which the court case is dealing with in part) one could argue that OpenAI is transforming works covered under copyright in a manner that isn't sufficiently transformative to pass the fair use exceptions to copyright and is thus committing copyright infringement. It's still not theft.
OpenAI doesn't "require stealing the world's books" in order to do what they do. Their product is vastly more effective and useful because it was trained on such a wide corpus of material, but likewise a xerox machine that won't make any copies of anything under copyright is vastly less useful and effective as one that will. Likewise a VCR that refuses to record from TV is less useful than one that will. A CD drive that refuses to rip MP3s from CDs is less useful than one that will. A BitTorrent client that refuses to send or receive items subject to copyright is less useful than a client that will send and receive those items. The fact that a product is better by it's ability to be usable in committing copyright infringement is neither evidence that the product itself is infringement, nor a strong argument in favor of preventing the product from being capable of such infringement.
You must have missed the NYT's suit and initial evidence. Not a problem, you'll fit right in with all the other temporarily embarrassed millionaires simping for this plundering of the commons..
2) The models explicitly don't have every book inside.
At best (which the court case is dealing with in part) one could argue that OpenAI is transforming works covered under copyright in a manner that isn't sufficiently transformative to pass the fair use exceptions to copyright and is thus committing copyright infringement. It's still not theft.
OpenAI doesn't "require stealing the world's books" in order to do what they do. Their product is vastly more effective and useful because it was trained on such a wide corpus of material, but likewise a xerox machine that won't make any copies of anything under copyright is vastly less useful and effective as one that will. Likewise a VCR that refuses to record from TV is less useful than one that will. A CD drive that refuses to rip MP3s from CDs is less useful than one that will. A BitTorrent client that refuses to send or receive items subject to copyright is less useful than a client that will send and receive those items. The fact that a product is better by it's ability to be usable in committing copyright infringement is neither evidence that the product itself is infringement, nor a strong argument in favor of preventing the product from being capable of such infringement.