I also quite like "AGI safety from first principles": https://www.lesswrong.com/s/mzgtmmTKKn5MuCzFJ