i bet you could get 99% of the performance with less powerful LLMs

→ View original post on X — @garymarcus