Josh Engels

Engels studies whether advanced AI systems can be understood and kept under control. He co-authored influential tests of sparse autoencoders—tools for identifying internal model features—and scaling laws for supervising systems that may outperform their overseers.