A checklist format of what we need to succeed at AI Safety, and it’s specifically safety and not alignment as there’s a pretty good case for a Control agenda. I care a lot about existential threats to humanity or to other things we value, but I also think that the upside from AGI is at least comparable. Mandatory