Matthew
@matthew
I want to talk to ChatGPT about books—granularly reference, make outlines, debate core points, expand on topics, etc. I would easily pay $20 per book to do this. Why can't OpenAI facilitate that txn, and why wouldn't authors be open to it? The alternative (pirating) feels like limewire before iTunes came around.
9 replies
5 recasts
25 reactions
JB Rubinovitz
@rubinovitz
I don't think publishers will ever trust ai companies with their books after all the copyright infringement. May be an interesting use case for crypto x ai but I dont think the market is big enough for the trouble.
1 reply
0 recast
0 reaction
Matthew
@matthew
I agree that there's a level of trust that needs to be built and hasn't until now. But the same was true for music, and that didn't stop Steve Jobs. Separately, how does crypto fit into it?
1 reply
0 recast
0 reaction
JB Rubinovitz
@rubinovitz
you could provably enforce data rights with crypto.
2 replies
0 recast
0 reaction
Matthew
@matthew
So I can prove that I interacted with the Kindle contract to buy a specific book, and that gives me access to the file?
1 reply
0 recast
0 reaction
JB Rubinovitz
@rubinovitz
On the model pipeline level I'd imagine provably training a model on book data without actually giving anyone else access to that data. And then you could provably pay the publisher whenever that model is used.
1 reply
0 recast
0 reaction
Matthew
@matthew
Well presumably they'd need to "give" OAI the transcripts just as record labels had to give MP3s to Apple so they could sell them. But really the helpful part is importing the book text into the LLM's context and generating embeddings, and maybe some quick start prompts.
2 replies
0 recast
0 reaction
Matthew
@matthew
I would imagine if I'm the publisher, the one thing I would *not* want ChatGPT users to be able to do is to export a copy of the transcript in a few clicks. Which OpenAI could definitely do given their control of both the model and the interface.
0 reply
0 recast
0 reaction
