Cost per task & time per task really need to be the measurements going forward
I’m convinced any model can produce brilliant outputs given enough test-time compute
Give me unlimited time and I might have a sliver of a chance at world-championship chess, but that’s not the setup, for a reason
顯示更多