Now that we see how poorly scientific literature AIs work we can understand how screwed the PDF-based traditional science publishing system is. Imagine the AIs we would have for science had we created an open, minable, free infrastructure as proposed by @brembs.mastodon.social.ap.brid.gy