everyone likes to talk about exponential gain of AI capabilities.... where's the corresponding talk about exponential demand -> exponential infrastructure to host it all?
personally one of my answers is we'll have liquid models where they fluidly downscale into child models who run locally on consumer hardware hybrid back to cloud as needed another of my answers is that the integrations & social adjustments will move slower than ppl think
More notes on nearby ideas
Tweet · Nov 3, 2023what's your take on AI capability "ceiling" and the shape of the curve from here to there / what is needed to get there?Reply · Aug 18, 2022@ask_ourself The other factor is that I don’t believe it will be all-or-nothing immediate Foom, because I think hardware constraints are a rate limit…Reply · Apr 19, 2024@RokoMijic @ESYudkowsky @robinhanson i remember debating w you last year about infrastructure & takeoff speed - good for you to update on evidence! 🫶…Tweet · Feb 4, 2025i feel like i need a better understanding of physical constraints on ai (power, DCs, supply chain, even fiber ig) ... there must be someone who's done…