Post

cpu is dogshit slow for inference, and self-replicating worms already exist (you may know them as rVPNs or botnets) alas they're just repackaging an existing problem as "you should be afraid of everyone's ai except ours"

I will eat my hat if a smart tv can host enough inference to devise *new* attacks rather than just exploit the same existing attacks and just happen to bring along an additional giant multi-gigabyte payload for the hell of it, which definitely lowers reproduction fitness.

my macbook air barely manages to run a 27b parameter model, and that shit ain't capable of launching new attacks that an enterprising North Korean isn't already capable of. let along a dumb router or IoT shit device

no, they're specifically fearmongering about open models

detecting stolen api keys is more or less a solvable problem, they're claiming there would be no central point of control to detect and stop these