From the point of AI safety, slow progress seems to be beneficial.