When Can Proxies Improve the Sample Complexity of Preference Learning? | Read Paper on Bytez