They say it can increase your productivity, and hot damn does that man need it. LLMs work best when manipulating input text along with contextual content. That sounds almost exactly like partially-edited manuscript and conceptual thought doodles plus several existing storyline books.
Admit it, if he dies before finishing, and someone publishes the result of such an AI output, you would read it. So he might as well get some AI help to finish it “himself”, sooner, right?
Unfortunately BrandoSando has already said he wouldn’t do it because their writing styles are too different.
Also the dude has writing plans for the next 30 or so years mapped out.
IIRC, GRRM has also said that he does not want any other author to finish ASOIF posthumously if he dies before completing it