Facts aren't copyrightable. Expression is. LLMs reproduce expression from the works they were trained on. The way they are being trained involves making an unlicensed reproduction of works. Both of those are pretty straightforwardly infringement of an exclusive right.

Establishing an affirmative defense that it's transformative fair use would hopefully be an uphill battle, given that it's commercial, using the whole work, and has a detrimental effect on the market for the work.