urbanriot Here’s a simple example for this situation. Keep in mind that it’s entirely hypothetical; not a commentary about you or a claim that I have actually written a book.
Let’s say that I wrote a detailed, well-respected book explaining the intricacies of modern system configurations, including obscure options and their interactions. This book took a huge amount of research and testing, and countless hours getting the formatting, layout wording, descriptions just right. My 375-page tome took a considerable effort, in some cases, taking time off work to get it published in time.
Without compensation or permission, this is part of a dataset that the people of Google download and feed to their model. When you issue your questions, it spits out an answer that you use to solve your problem. You’re happy, because it saved you some time. Ordinarily, being the dedicated professional that you are, you might have ordered that book to make sure you were up to speed on all of the latest configuration issues; now you don’t have to. That’s one copy I won’t be seeing royalties from. How many times does this happen?
I wrote the book with the intention of being helpful, and at least being fairly compensated for the sizeable effort, not to make money for Google while they give me nothing. When their model ingests my content, that should be the basis of an agreement between us that they will be paying me for the repeated use of that content, especially since I will be selling fewer copies of the work specifically because their model ingested it and is generating answers based on my work. It’s not greed; it’s how the economy works. If we allow them to break that model, who will be inclined to produce these essentially human works that disseminate knowledge and improve the human condition?
It’s also a question of giving credit where it’s due. If I’m a post-doctoral physicist, and I produce a paper that depends on many concepts, I don’t describe the concepts of gravity in the introduction as if I discovered it, or present my mathematical results as if I’m introducing the world to differential calculus. This has been done, and proper citations are expected. If I use some sort of proof, I credit the person or people whose research I used as source material, and I don’t reproduce their work, slightly re-worded, as if it’s my original work, depriving the journal in which their work was published of sales and their research of its due recognition.
I can’t and shouldn’t do this, and we should not even be considering the idea of an exception for LLMs, which are ultimately intended to turn a profit for their developers. How would they feel if we said that, since they didn’t pay for training material, they are also not allowed to charge any fees for use of the models, in spite of the staggering electricity bills that go along with operating them? Free stuff in, free stuff out, no?