Our project furthers the progress of Scale Oversight through automation of the sandwiching paradigm. In the Bowman et al. (2022) paper, the question is presented of how humans can effectively prompt unreliable, superhuman AIs to answer ques...