.@MaxNadeau_ on why the Whac-A-Mole approach frontier labs take to aligning AIs is not going to work for superintelligence: "I'm not that enthusiastic about the Whac-A-Mole approach that the frontier labs take to aligning AIs as a technique for aligning superintelligence. I think that we should not bet on that." "It's useful for aligning the AI systems of today, and I think that's valuable because the AI systems of today can be a big help in making further progress on alignment techniques…