6 ms·
In the general case, yes, but they can verifiably reproduce at least some copyrighted works verbatim, which implies, at the minimum, that their content is store
by int_19h 2y ago
In the general case, yes, but they can verifiably reproduce at least some copyrighted works verbatim, which implies, at the minimum, that their content is stored in model weights in some fashion.
- lsaferite 2y agoIt implies that the token procession probability was unique enough that with a low entropy token stream and the proper starting token stream you could recreate portions of the original content *strictly based on probabilities*.
- johanyc 2y agoEveryone knows the training data is stored in some way in the LLM. The point is the use of the copyrighted material is transformative. Remember google books, it literally shows photocopy of pages of books but the court ruled it’s fair use. A simplified explanation is book vs search engine and book vs ai chatbot are very different from each other.
- SideburnsOfDoom 2y ago"a photocopy of pages of books" is exactly that, pages of an existing book. It doesn't pretend to be something else. The output of a LLM, when based heavily on that same page, pretends to be something novel. IMadeThis_Meme.gif