look, there is some idea here. stable diffusion & gpt are extremely gpu intensive to run locally, so only accessible to the few. the central hubs censor the fun content. what’s that thing that’s censorship-resistant and allows us to coordinate resource-intensive compute again?