Seems like a good time to revisit Eliezer's prediction from 13 years ago. Interesting to see that he was uncertain about "what signs of competence" self-improving AI would exhibit, how long takeoff would take, and what methods would start the feedback loop.