@bedirthan @jalefkowit in their case study, they needed to hack a dozen smart devices and chain them together to run a tiny, dumb model not capable of writing coherent computer code at 0.39 tokens/second. By contrast, a company serving a model at 50 tokens/second is considered "slow" for many long running agentic tasks. They estimate days before such a virus could duplicate to the next set of machines, all the while your fridge and TV and printer would laggy/unresponsive while this happened