I’ll say this. These things are really weird — look at the hugging face transcripts — and they seem to be capable of things like solving Millennium math problems. There’s no guarantee that they keep getting smarter or better able to affect the real world, but they’re already extremely good at hacking, and all the predictions about slowing down haven’t happened yet. Moreover, at a minimum, there are real worries that these things may be able to empower bad actors with respect to hacking or, significantly more serious, biowarfare. I would say the smart version of the argument is that this is a leap into the unknown, and nobody’s sure where it ends up. Saying p(doom) = .1 or whatever is just a way of expressing this uncertainty.