The annoying thing about the article is the hand wringing. If anything this is the least dubious way to acquire training material. They are paying the correct party to get that material. AI training is at least as transformative as throwing the words into an inverted index for web search.

One thing this demonstrates is that there's still training material to be mined that's original, and high enough quality to have been published on paper. This source of material won't get mined out until all the similar material originally written in major foreign languages get scanned and translated for training purposes. We are in the gentle slopes of the lower foothills of the never gonna be AGI so I'll call it singularity.

Otherwise, I think it's good that books are being scanned. Please, AI companies, keep all of the scans, clean them so they can be compressed tightly, back them up to an insane degree, encourage your employees to make their own archives and share them with their friends, and leak them to libgen, annas-archive, etc.

https://annas-archive.gl/donate