I'm not an expert in LLM training, but I think we can all agree that the writing on the Internet is generally very low-quality and surface level compared to the depth of books. Most books in the past were even edited by a separate person from the writer!
Granted that that is indeed the case, surely the quantity of high-quality writing available online (including things like Project Gutenberg) still dwarfs the quantity remaining in obscure unscanned books.
Humans manage to be pretty intelligent with only reading perhaps a few thousand books in their lifetime. It seems unlikely that AGI will appear but only once it has read that last out of print 1983 book on knitting patterns.