In today's episode of WGMI: We're Gonna Make It, we delve into the future of AI, focusing on the challenge of aligning superhuman models. Discover the intricacies of weak-to-strong generalization and explore the methodologies for supervising AI models beyond human capabilities. We discuss the importance of understanding AI's potential to mimic supervisor mistakes and the implications of pretraining leakage. Join us as we outline key future research directions and the necessity of establishing reliable AI alignment methods. Tune in to grasp the complexities of AI superalignment and the steps toward ensuring these powerful models align with human values.