Well - there's nothing left to argue about - I do believe we have bigger climate killers than large computing centers, but it is a worrying trend to spend that much energy for an investment bubble on what is essentially an somewhat advanced word prediction. However, if we could somehow get the wish.com version of Tony Stark and other evil pisswads to die out, then yes, using LLMs for some creative ideas is a possibility. Or for references to other sources that you can then check.

However, the way those models are being trained is aimed at impressing naive people and that's very dangerous, because those people mistake impressively coherent sentences for understanding and are willing to talk about automating tasks upon which lives depend.