I really enjoyed CFAR ten years ago, it was what finally got me to take action and finally start properly studying how to contribute to AI capabilities. I now think that was a first-order bad thing for it to have done, and it’s unclear whether my response to realizing that was bad is remotely close to undoing the p(doom|action)-p(doom|counterfactual inaction) change. I think CFAR has promising ideas for how to think well but I generally think most ways of making people better end up mostly impacting the world by amplifying AI capabilities researchers who are least doom avoidant in their approach, and it’s not clear to me how to prevent that given that CFAR was already at the time designed to get people to think about consequences.
I really enjoyed CFAR ten years ago, it was what finally got me to take action and finally start properly studying how to contribute to AI capabilities. I now think that was a first-order bad thing for it to have done, and it’s unclear whether my response to realizing that was bad is remotely close to undoing the p(doom|action)-p(doom|counterfactual inaction) change. I think CFAR has promising ideas for how to think well but I generally think most ways of making people better end up mostly impacting the world by amplifying AI capabilities researchers who are least doom avoidant in their approach, and it’s not clear to me how to prevent that given that CFAR was already at the time designed to get people to think about consequences.