The problem is that even if we were to assume AI doomers are completely right, the current approach to AI "safety" is absolutely, completely powerless to do any

The problem is that even if we were to assume AI doomers are completely right, the current approach to AI "safety" is absolutely, completely powerless to do anything that would actually matter to stop it or make it safer. Writing long articles about the theoretical threats or the ethics of AI is fine, but to call it AI safety is misleading. At best, you can censor the output of the models but imo that has nothing to do with "safety".