Eliezer Yudkowsky’s views on AI safety


Topic View
AI timelines see here for 2083 as Carl’s estimate, and in IEM Eliezer says “I’m currently trying to sort out with Carl Shulman why my median is forty-five years in advance of his median” (p. 83) so that puts Eliezer’s estimate as of 2013 at around 2038. See also this thread and this tweet and this comment.
Value of decision theory work
Value of highly reliable agent design work
Difficulty of AI alignment See this tweet.
Shape of takeoff/discontinuities in progress
Type of AI safety work most endorsed
How “prosaic” AI will be
Kind of AGI we will have first (de novo, neuromorphic, WBE, etc.)
Difficulty of philosophy here is one remark