Yeah, I find the whole concept a bit of a nonstarter until models that I can run on a single somewhat-normal-consumerish machine (e.g. a Mac Studio) with decent

Yeah, I find the whole concept a bit of a nonstarter until models that I can run on a single somewhat-normal-consumerish machine (e.g. a Mac Studio) with decent capability and speed have appeared. I’m not interested in sending literally everything across the wire to somebody else’s computers, and unless the AI bubble pops and cheap GPUs start raining down on us I’m not interested in building some ridiculous tower/rackmount thing to facilitate it either.