We present a new research direction for superalignment, together with promising initial results: can we leverage the generalization properties of deep learning to control strong models with weak supervisors? Written by: Elis Wanyama Posted on: April 19, 2024