one thing I wish I understood better is the actual cost of inference and what it'll look like in the future as supply has time to catch up with demand and models are further optimized per use case... rn pricing is like early days of Uber or the scooter craze, who knows