What’s been happening in AI alignment?

In this talk, Rohin surveys conceptual progress in AI alignment over the last two years. While the talk assumes basic knowledge of the arguments for AI risk, technical knowledge is not necessary.