Further reading: The rest of the courses' slides, the works of Dan Hendrycks and Collin Burns in particular and this short blog post I made on the offense-defense model in AI safety a while back.