Berlatsky is right about this danger, but the challenge of talking about LLMs is that there are multiple domains of insurmountable problems with them, each sufficient to be "the danger" of LLMs. I'll list a few here, each of which could be a long essay -- and I'll get to that when I can.
Noah Berlatskyagain, the danger from AI is not that it will attain sentience and destroy us. the danger is that it's a mindless flattery machine and being mindlessly flattered literally drives people insane.