Theoretical guarantees on the best-of-n alignment policy | Read Paper on Bytez