On scalable oversight with weak LLMs judging strong LLMs | Read Paper on Bytez