Aren’t books massively outweighed by the crawled internet corpus?

I would doubt that because books are probably weighed as higher quality and more trustworthy than random Reddit posts

Especially if it's unsupervised training