The people who wrote AI 2027, the scenario about how AI could kill us all if things keep going at this speed, just released a new scenario about how to make AI go really well for humanity. TLDR: The U.S. leads an effort to delay superintelligence until 2040, make AI research much more public and transparent, let many companies around the world catch up to the frontier, and build datacenters in deliberately vulnerable locations so the compute can be destroyed if the deal breaks down and the rac...